Redemption recommendation method and device, equipment, storage medium and product

Abstract

The invention relates to the technical field of computers, and discloses an exchange recommendation method and device, equipment, a storage medium and a product, and the method comprises the steps: obtaining a user portrait of a user and a dynamic value coefficient of each target article in response to a user recommendation request triggered by the user, the target articles being commodities exchanged by using points; determining a context feature corresponding to the user recommendation request, inputting the user portrait, the dynamic value coefficient and the context feature into a fusion sorting model, and outputting a final recommendation score of the target article; and generating an exchange recommendation list according to the final recommendation score. By integrating the user portrait, the dynamic value coefficient of the target article and the context feature corresponding to the request, the recommendation result not only can fit the long-term preference of the user, but also can adapt to the current demand scene of the user and the real-time value of the article, so that the individuation and accuracy of exchange recommendation are remarkably improved, and the user experience is improved. And the acceptance of the user to the recommendation list and the credit exchange conversion rate are improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a redemption recommendation method, apparatus, device, storage medium, and product. Background Technology

[0002] Existing points redemption recommendation technologies generally employ static rules, treating points as a fixed-value currency and recommending items according to fixed categories or point rankings. For example, they might display products that can be redeemed with current points, arranged from lowest to highest or highest to lowest point requirement. This results in recommended products failing to meet users' unique preferences, leading to homogenized recommendations. Furthermore, by ignoring the dynamic fluctuations in points value, a large number of user points are redeemed arbitrarily, resulting in low redemption efficiency and reduced user engagement and satisfaction with the points system. Summary of the Invention

[0003] The main purpose of this application is to provide a redemption recommendation method, apparatus, device, storage medium, and product, aiming to solve the technical problem that existing points redemption recommendation technologies result in homogenized recommendations, reducing users' sense of participation and satisfaction with the points system.

[0004] To achieve the above objectives, this application proposes an exchange recommendation method, the method comprising: In response to a user-triggered user recommendation request, the system obtains the user profile of the user and the dynamic value coefficient of each target item, wherein the target items are goods redeemed using points. Determine the contextual features corresponding to the user recommendation request, and input the user profile, the dynamic value coefficient, and the contextual features into the fusion ranking model to output the final recommendation score of the target item; A redemption recommendation list is generated based on the final recommendation score.

[0005] Optionally, the step of obtaining the user's user profile and the dynamic value coefficient of each target item in response to a user-triggered user recommendation request includes: Based on the user recommendation request, the user profile corresponding to the user is extracted from the user feature library, and the user profile is a feature vector; Obtain real-time attribute information and score information for each target item, and calculate the dynamic value coefficient of each target item based on the real-time attribute information and the score information.

[0006] Optionally, before the step of extracting the user profile corresponding to the user from the user feature library based on the user recommendation request, the method further includes: Collect multi-source heterogeneous data corresponding to the user; The multi-source heterogeneous data is preprocessed and the preprocessed data is converted into features to generate structured feature data, which includes user behavior features, user points features and item attribute features. The user profile corresponding to the user is updated based on the feature data, and the user profile is stored in the user feature database.

[0007] Optionally, the step of obtaining real-time attribute information and score information of each target item, and calculating the dynamic value coefficient of each target item based on the real-time attribute information and the score information, includes: Obtain real-time attribute information and points information for each target item, and determine calculation factors based on the real-time attribute information. The calculation factors include cost factors, inventory pressure coefficients, and marketing adjustment factors. The value of the goods corresponding to the points information is determined according to the preset exchange ratio. The calculation factor, the integral information, and the commodity value information are input into a preset calculation model to calculate the dynamic value coefficient of each target item.

[0008] Optionally, the fusion ranking model includes a linear model and a deep model; The step of determining the contextual features corresponding to the user recommendation request, inputting the user profile, the dynamic value coefficient, and the contextual features into the fusion ranking model, and outputting the final recommendation score of the target item includes: The first feature of each of the target items is determined based on the linear model; Determine the contextual characteristics of the user when the user recommendation request is triggered; The user profile, the dynamic value coefficient, and the contextual features are input into the deep model, and the second feature of the target item is output. The final recommended score for the target item is determined based on the first feature and the second feature.

[0009] Optionally, after the step of generating the redemption recommendation list based on the final recommendation score, the method further includes: Collect user interaction data on target items in the redemption recommendation list; The interactive behavior data is labeled to generate training samples, which include positive samples and negative samples. The positive samples are the target items that the user redeemed, and the negative samples are the target items that the user did not redeem or click on. The fusion ranking model is updated using the training samples with the goal of minimizing the cross-entropy loss function.

[0010] Furthermore, to achieve the above objectives, this application also proposes an exchange recommendation device, which includes: The feature acquisition module is used to respond to a user recommendation request triggered by a user, and to acquire the user profile of the user and the dynamic value coefficient of each target item, wherein the target items are goods redeemed using points; The scoring calculation module is used to determine the context features corresponding to the user recommendation request, and input the user profile, the dynamic value coefficient and the context features into the fusion ranking model to output the final recommendation score of the target item; The redemption recommendation module is used to generate a redemption recommendation list based on the final recommendation score.

[0011] In addition, to achieve the above objectives, this application also proposes an exchange recommendation device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the exchange recommendation method as described above.

[0012] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and which, when executed by a processor, implements the steps of the exchange recommendation method described above.

[0013] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the redemption recommendation method described above.

[0014] This application discloses a method for responding to user-triggered recommendation requests, obtaining the user profile and dynamic value coefficients of each target item (items redeemable using points); determining the contextual features corresponding to the user recommendation request; inputting the user profile, dynamic value coefficients, and contextual features into a fusion ranking model; and outputting a final recommendation score for the target item. A redemption recommendation list is generated based on the final recommendation score. By integrating the user profile, target item dynamic value coefficients, and contextual features corresponding to the request, the recommendation results align with both long-term user preferences and current user needs and item real-time value, significantly improving the personalization and accuracy of redemption recommendations. This helps increase user acceptance of the recommendation list and the points redemption conversion rate. In actual testing, compared to static rule-based recommendations, this solution increased the points redemption conversion rate by 25.2% to 30.4%. Simultaneously, the recommendation click-through rate increased by 18.3% to 22.1%, indicating a significant increase in user acceptance and engagement with the recommendation results. Users no longer need to search through thousands of items; the system directly pushes their most likely favorite and currently most cost-effective options, simplifying the decision-making process. Furthermore, the platform can integrate marketing objectives (such as clearing specific inventory or promoting new products) into dynamic value assessment. By adjusting parameters, it can intelligently guide users to target products, thereby achieving precise operations. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the redemption recommendation method of this application; Figure 2 This is a flowchart of the recommended redemption method for this application; Figure 3 This is a flowchart illustrating the second embodiment of the redemption recommendation method of this application; Figure 4 This is a structural diagram of the redemption recommendation system for this application; Figure 5 This is a flowchart illustrating the third embodiment of the redemption recommendation method in this application; Figure 6 This is a schematic diagram of the module structure of the redemption recommendation device in an embodiment of this application; Figure 7 This is a schematic diagram of the device structure of the hardware operating environment involved in the redemption recommendation method in this application embodiment.

[0018] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0020] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0021] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, feature fusion, and program execution functions, such as a computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions. The following description uses an intelligent redemption recommendation system as an example to illustrate this embodiment and the subsequent embodiments.

[0022] Based on this, the embodiments of this application provide a redemption recommendation method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the redemption recommendation method of this application.

[0023] In this embodiment, the redemption recommendation method includes: Step S10: In response to a user recommendation request triggered by a user, obtain the user profile of the user and the dynamic value coefficient of each target item, wherein the target item is a product redeemed using points.

[0024] It should be noted that user recommendation requests can be user actions triggered by the user within a points redemption platform (such as an app or mini-program) to obtain recommendation services. For example, the system automatically triggers a recommendation command when the user clicks the "Points Redemption Recommendation" entry or enters the redemption page. User profiles are high-dimensional feature vectors constructed based on multi-source user data (basic attributes, behavioral records, points status, etc.) to comprehensively represent users' personalized preferences, such as category preferences and points usage habits.

[0025] Understandably, the target item refers to physical or virtual goods on the platform that users can redeem using points, such as digital products and coupons. The dynamic value coefficient is a quantitative parameter that is calculated in real time and reflects the current cost-effectiveness of redeeming the target item with points. It changes dynamically with factors such as item inventory, marketing activities, and costs. The higher the value, the more cost-effective it is to redeem the item with points at present.

[0026] Understandably, when the intelligent redemption recommendation system detects a user recommendation request triggered by a user, it first extracts the user profile corresponding to that user from the preset user feature database based on the user identifier carried in the request, such as the user ID. At the same time, the system obtains relevant data of all target items that can be redeemed with points in real time, and calculates the dynamic value coefficient of each target item through a preset calculation model. Of course, it can also call the pre-calculated dynamic value coefficient from the real-time running dynamic value evaluation module.

[0027] Step S20: Determine the contextual features corresponding to the user recommendation request, and input the user profile, the dynamic value coefficient, and the contextual features into the fusion ranking model to output the final recommendation score of the target item.

[0028] It should be noted that contextual features refer to real-time scenario-related data when a user triggers a user recommendation request. This can include the time information of the request (such as time period, weekday / weekend), the type of device the user logged in with (such as mobile, PC), whether it is currently a holiday or a special event period for platform points redemption, and the regional marketing rules associated with the user's current geographical location. These features are used to characterize the specificity of the user's current redemption scenario and make up for the differences in real-time needs that cannot be covered by relying solely on the user's historical profile.

[0029] As can be understood, the fusion ranking model refers to a recommendation model that combines linear and deep models. The linear model focuses on capturing explicit, rule-based feature relationships, such as the direct matching of user age segments with target items in the maternal and infant category. The deep neural sub-model focuses on uncovering implicit, complex, multi-dimensional feature relationships, such as the potential association between holiday scenarios and snack items. The final recommendation score is a quantitative result output by the fusion ranking model after integrating user profiles, dynamic value coefficients, and contextual features, used to directly represent the degree of fit between the target item and the user's current redemption needs.

[0030] Specifically, the intelligent redemption recommendation system integrates the metadata of user recommendation requests to form the contextual features corresponding to the request; then, it performs unified preprocessing on the user profile, the dynamic value coefficients (continuous values) of each target item, and the aforementioned contextual features; it converts all categorical features into low-dimensional dense vectors through an embedding layer, and normalizes all continuous features to map them to a preset numerical range to ensure that the feature format meets the model input requirements; finally, it inputs the preprocessed three types of features into a fusion ranking model, and the outputs of the linear model and the deep model are finally merged through a weighted sum (or concatenated and then through a linear output layer) to generate the final recommendation score.

[0031] Step S30: Generate an exchange recommendation list based on the final recommendation score.

[0032] It should be noted that the redemption recommendation list is a collection of redeemable target items displayed to users based on the final recommendation score. It is presented in list form and includes key information such as item name, redemption price in points, and item image.

[0033] It should be understood that the intelligent redemption recommendation system first sorts all target items in descending order of their final recommendation scores. If there are items with the same score, the dynamic value coefficient of the target items is used as the basis for a second sort, prioritizing items with higher dynamic value coefficients to ensure the uniqueness of the sorting logic. Next, the sorted items are further filtered using preset filtering rules to remove target items with zero current inventory, expired redemption activities, or insufficient user points balance, avoiding recommending invalid options to users. Subsequently, the list display format is adapted according to the user's login device type. For example, mobile devices prioritize displaying concise information, such as item images, point prices, and key selling points, while PC devices can supplement with dynamic value coefficient interpretations, such as current cost-effectiveness being 15% higher than similar items. Finally, the processed item sequence is integrated into a structured redemption recommendation list and pushed to the user's currently accessed interface (such as the points redemption homepage, recommendation zone, etc.) via API interface, completing the list generation and display.

[0034] Furthermore, to implement a closed-loop update mechanism based on real interactive feedback, enabling the model to continuously learn the latest user preferences and dynamically optimize the linear model weights and deep model network parameters, the recommendation capability achieves self-evolution, ensuring the stability and adaptability of long-term recommendation performance and preventing the model from failing due to changes in user needs. Following step S30, the following is also included: Collect user interaction data on target items in the redemption recommendation list; label the interaction data to generate training samples, which include positive and negative samples, where positive samples are target items redeemed by the user and negative samples are target items not redeemed or clicked by the user; update the fusion ranking model using the training samples with the goal of minimizing the cross-entropy loss function.

[0035] It should be noted that interaction behavior data refers to all operation records generated by users while viewing the redemption recommendation list, including but not limited to clicking on target items, dwell time, adding to redemption favorites, initiating redemption (success / failure), canceling redemption, etc., which directly reflect users' true preferences for recommended items. Training samples are model training data units formed by labeling interaction behavior data. Positive samples refer to target items that users ultimately redeemed, while negative samples refer to target items that users did not click on or clicked but did not redeem. Negative samples need to exclude non-preference factors such as insufficient inventory or insufficient points that prevented redemption.

[0036] Additionally, it should be noted that the cross-entropy loss function is used to measure the deviation between the recommendation score output by the fusion ranking model and the user's actual interaction behavior. In the recommendation ranking scenario, it can effectively quantify the degree of deviation of the model from the goal of ranking user-preferred items higher.

[0037] It should be understood that during the training or updating of the fusion ranking model, the training data can be samples constructed from the feedback log library. Positive samples are the products that users ultimately redeem, and negative samples are the products that were exposed but not clicked or redeemed. The optimization objective can be to minimize the cross-entropy loss function, with the goal of making the user redemption probability predicted by the model consistent with the actual behavior. At the same time, historical data is used for offline training in the early stage of training, and incremental training and model updates can be performed regularly (e.g., daily) using incremental data after the system goes online.

[0038] Specifically, after the system outputs the redemption recommendation list, it can capture user interaction data in real time through front-end tracking and log collection tools. Simultaneously, it stores the user's profile data, contextual features, and the dynamic value coefficient of the corresponding target item, ensuring data traceability to the specific recommendation scenario. Secondly, the collected interaction data is labeled. Successfully redeemed target items and their associated feature data are labeled as positive samples. From items not clicked or clicked but not redeemed, target items with sufficient stock and enough user points for redemption are selected, and their associated feature data are labeled as negative samples. The number of positive and negative samples is balanced at a preset ratio (e.g., 1:5) to avoid sample imbalance affecting model training. Finally, the labeled training samples are input into the fusion ranking model. With minimizing the cross-entropy loss function as the optimization objective, the gradient change of model parameters is calculated using the backpropagation algorithm. The parameters are iteratively updated at a preset learning rate until the loss function value converges to a preset threshold or reaches the maximum number of iterations, completing one model update. This allows the model to output recommendation scores that more accurately match user preferences in subsequent iterations.

[0039] In one example, reference Figure 2 , Figure 2This is a flowchart of the recommendation method used in this application. First, the data acquisition layer is responsible for collecting multi-source heterogeneous data from various business databases, log servers, and real-time data streams (such as Kafka). After collection, data cleaning, preprocessing, and feature extraction are performed. Offline / real-time user profile updates are executed, and the updated user profiles are stored in the user feature library. Simultaneously, the dynamic value assessment model runs in real-time. When the system receives a user recommendation request, it obtains the user's multi-dimensional profile feature vector U and the candidate product set and its real-time dynamic value coefficient V. Then, it constructs the model input features, fuses (U, V, context C), and inputs them into the intelligent recommendation model, which calculates the recommendation score for each product. The candidate products are then sorted in descending order according to their scores, generating and returning a Top-N personalized recommendation list. After the recommendation is completed, the recommendation result log is recorded, and user feedback on the recommendation list is monitored and collected. User profiles are updated offline or in real-time based on the feedback.

[0040] In this embodiment, in response to a user-triggered recommendation request, the system acquires the user profile and the dynamic value coefficients of each target item, where the target items are goods redeemed using points. It determines the contextual features corresponding to the user recommendation request and inputs the user profile, dynamic value coefficients, and contextual features into a fusion ranking model, outputting the final recommendation score for the target item. A redemption recommendation list is generated based on the final recommendation score. By integrating the user profile, the dynamic value coefficients of the target items, and the contextual features corresponding to the request, the recommendation results align with both long-term user preferences and current user needs and the real-time value of the items, significantly improving the personalization and accuracy of redemption recommendations. This helps increase user acceptance of the recommendation list and the points redemption conversion rate. Furthermore, the platform can integrate marketing objectives (such as clearing specific inventory or promoting new products) into the dynamic value assessment, intelligently guiding users to target products by adjusting parameters, achieving precise operation.

[0041] Reference Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the redemption recommendation method of this application. Based on the first embodiment described above, a second embodiment of the redemption recommendation method of this application is proposed. In the second embodiment, step S10 includes: Step S101: Based on the user recommendation request, extract the user profile corresponding to the user from the user feature library, wherein the user profile is a feature vector.

[0042] Understandably, when the system receives a user recommendation request, it can extract the user's unique identifier (such as an encrypted user ID) from the request parameters and use this identifier to construct the search conditions for the user feature database. Then, the system calls the query interface of the user feature database to accurately match and extract the complete profile data corresponding to the user based on the identifier. The data can be returned in the form of a high-dimensional feature vector with preset dimensions. For example, the first dimension of the vector is the age segmentation code, the second dimension is the normalized value of the points balance, and the third dimension is the category preference weight, etc.

[0043] Furthermore, to enable user profiles to be dynamically updated as user data changes, the degree to which user profiles accurately reflect actual user characteristics is improved. Before step S101, the following steps are also included: Collect multi-source heterogeneous data corresponding to the user; preprocess the multi-source heterogeneous data and convert the preprocessed data into features to generate structured feature data, which includes user behavior features, user points features and item attribute features; update the user profile corresponding to the user based on the feature data and store the user profile in the user feature database.

[0044] It should be noted that multi-source heterogeneous data can be user-related data collected from various business databases, log servers, and real-time data streams (such as Kafka), including user points, user behavior logs, item management systems, and user basic information systems. Structured feature data refers to converting preprocessed data into a standardized, computable feature format. Among these features, user behavior features (such as the number of redemptions in the past 30 days and the click-through rate of preferred categories) reflect user behavior habits, user point features (such as point balance, point validity period, and historical consumption rate) reflect user point status, and item attribute features (such as item category code, the quantity required for point redemption, and whether the item has participated in marketing activities) reflect the core attributes of the target item.

[0045] Understandably, the system can collect user behavior logs through front-end tracking tools, collect user points data by calling the points system API, and collect target item attribute data through the item management system interface. Simultaneously, it extracts basic data such as user registration location and age segments from the user basic information database, completing the comprehensive collection of multi-source heterogeneous data. Secondly, the collected data undergoes preprocessing: abnormal data in the behavior logs is removed, such as records of the same item being clicked more than 10 times within one minute; missing values ​​in the points data are supplemented, such as filling missing daily data with the user's average points earned over the past 3 months; and date and numerical data of different formats are uniformly converted to a preset format. Next, the preprocessed data is converted into structured features, and behavioral features such as "redeem frequency in the past 30 days" and "preference category weight" are calculated for user behavior data. For example, points data is normalized to generate points features, and item attribute data is encoded to generate item attribute features. Finally, based on the user's unique identifier, the system locates the user's original profile in the user feature database, updates the corresponding dimensions in the profile with the newly generated structured feature data, and creates a completely new profile vector if the user is new. After the update, the profile is stored in the user feature database to ensure the data is up-to-date for subsequent retrieval.

[0046] Step S102: Obtain the real-time attribute information and score information of each target item, and calculate the dynamic value coefficient of each target item based on the real-time attribute information and the score information.

[0047] It should be noted that the real-time attribute information is dynamic data of the items synchronized in real time from the target item management system, marketing activity system, etc. Specifically, it includes the current inventory quantity of the target item, whether it participates in the points redemption special activity, etc., which are updated in real time as the item status changes; the points information is the number of points required to redeem the target item, as well as the platform's preset redemption rules corresponding to the points, such as whether 100 points are equivalent to 1 yuan cash value.

[0048] Furthermore, in order to consider not only the real-time attribute information of the target item (such as cost, inventory, marketing, etc.) when calculating the dynamic value coefficient, but also the commodity value information and points information determined by the preset exchange ratio, the dynamic value coefficient can comprehensively reflect the actual value of the item, operational needs, and the cost-effectiveness of points redemption. Step S102 may include: The system acquires real-time attribute information and points information for each target item, and determines calculation factors based on the real-time attribute information. The calculation factors include cost factors, inventory pressure coefficients, and marketing adjustment factors. The system determines the commodity value information corresponding to the points information according to a preset exchange ratio. The system inputs the calculation factors, the points information, and the commodity value information into a preset calculation model to calculate the dynamic value coefficient of each target item.

[0049] Understandably, the calculation factors are parameters that quantify the real-time characteristics of an item, including cost factors, inventory pressure coefficients, and marketing adjustment factors. The cost factor is the ratio of the item's cost price to its cash selling price, reflecting cost-effectiveness. The inventory pressure coefficient is the ratio of the item's current inventory to its safety stock threshold; the lower the inventory, the higher the coefficient, reflecting the need to clear inventory. The marketing adjustment factor is a bonus coefficient set according to the discount during the promotional period, and set to 1 during non-promotional periods. The preset exchange ratio refers to the platform's pre-set conversion relationship between points and cash, used to quantify points into cash value. Product value information refers to the cash value converted from points information through the preset exchange ratio. The preset calculation model is a mathematical formula that integrates calculation factors, points information, and product value information, used to transform multi-dimensional data into a single quantitative indicator of cost-effectiveness. Specifically, for each redeemable item i, at time t, its dynamic value coefficient is calculated in real time according to a preset calculation model. .in, = (Cash value of goods - Cost factor + Marketing adjustment factor) / (Points price * Inventory pressure coefficient). The higher the value, the more cost-effective it is to redeem the goods with points.

[0050] First, the system calls the target item management system in real time via API to obtain real-time attribute information of all redeemable items with points. Simultaneously, it extracts the corresponding points information for each item from the points rules system, i.e., the number of points required for redemption. Second, based on the real-time attribute information, three core factors are calculated: the cost factor equals the supplier's cost price divided by the cash selling price; the inventory pressure coefficient equals the current inventory divided by the safety stock threshold (set by the platform based on historical sales); and the marketing adjustment factor is set according to activity rules, such as 1.3 for discount activities, 1.2 for promotional activities, and 1 for no activities. Next, based on the platform's preset redemption ratio (e.g., 100 points equal 1 yuan), the product value information is calculated, converting the points cost into cash value for cost-effectiveness evaluation. Finally, the calculated factors, points information, and product value information are substituted into a preset calculation model, and the calculation results are normalized to obtain the dynamic value coefficient of each target item.

[0051] In one example, reference Figure 4 , Figure 4This is a structural block diagram of the redemption recommendation system of this application. The redemption recommendation method of this embodiment can be applied in the intelligent redemption recommendation system 100. Inside the intelligent redemption recommendation system 100, the data acquisition and preprocessing module 110 collects user, behavior, product, and points data through the data acquisition layer, and inputs it into the core computing engine 120 after data cleaning and feature engineering. The user profile modeling submodule 121 in the core computing engine 120 constructs user profiles offline based on the received data, the dynamic value assessment submodule 122 calculates the dynamic value coefficient of items in real time, and the intelligent recommendation algorithm submodule 123 generates recommendation logic based on the data and sends it to the database cluster 140. The user feature library, model database, and feedback log library of the database cluster 140 store the corresponding data respectively, providing support for the operation and iteration of each module. Next, the recommendation list generation submodule 131 of the recommendation and feedback interface module 130 completes the mixed sorting based on the data of the database cluster 140, pushes recommendations through the API interface, and collects user feedback by the feedback data collection submodule 132. It collects every interaction of the user with the recommendation results (such as exposure, click, redemption) and stores it in the feedback log library. This data is used for the continuous training and optimization of the subsequent model.

[0052] In this embodiment, a dynamic value coefficient is calculated based on the real-time attribute information and points information of the target item, ensuring that the dynamic value coefficient can reflect the current status and exchange value of the item in real time and reflect dynamic fluctuation characteristics.

[0053] Reference Figure 5 , Figure 5 This is a flowchart illustrating the third embodiment of the redemption recommendation method of this application. Based on the above embodiments, a third embodiment of the redemption recommendation method of this application is proposed. In the third embodiment, the fusion ranking model includes a linear model and a deep model, and step S20 includes: Step S201: Determine the first feature of each of the target items according to the linear model.

[0054] It should be noted that the linear model refers to the sub-model in the fusion ranking model responsible for capturing explicit and regularized feature associations. It processes input features through linear weighting formulas (such as logistic regression and linear regression) and outputs quantitative results with clear interpretability. It is suitable for mining direct and regularized matching relationships between users and target items, such as the strong association between users aged 25-35 and beauty-related target items. The first feature refers to the quantitative feature output by the linear model that represents the explicit matching degree between users and target items. Its value directly reflects the degree of fit based on explicit rules (such as user category preferences and matching of basic item attributes). It is different from the implicit features output by the deep model and has the characteristics of strong interpretability.

[0055] Understandably, the Wide part (linear model) in the fusion ranking model is a linear model responsible for memorizing, inputting cross features and binarized features. Among them, the cross features are used to memorize simple and direct rules such as "which age groups of users like which categories", such as user_age_bucket × category representing the combination of user age groups and product categories. The binarized features (such as is_product_in_promotion) can indicate whether the product is on promotion.

[0056] Specifically, explicit features suitable for the linear model can be selected from user profiles, target item attributes, and dynamic value coefficients. These include categorical features from user profiles, basic attributes of target items, and regularized indicators from dynamic value coefficients. Secondly, the selected features are preprocessed: categorical features are converted to one-hot encoding or numerical mapping, and continuous features are normalized to ensure the feature format is compatible with the linear model input. Next, the preprocessed features are input into the linear model, which performs a linear weighted summation of the features using preset weight parameters, while also introducing feature interaction terms. Finally, the weighted result is mapped to a preset interval through the output layer of the linear model (such as the Sigmoid function) to obtain the first feature corresponding to each target item. The higher the feature value, the higher the matching degree between the user and the target item based on explicit rules.

[0057] Step S202: Determine the contextual features of the user when triggering the user recommendation request.

[0058] It is understandable that determining the contextual features of a user when triggering the user recommendation request can be achieved by extracting the trigger time and login device identifier from the recommendation request metadata, combining this with the system's holiday interface to determine whether it is an activity period, and integrating these into features that include time and device dimensions; alternatively, it can be based on obtaining the user's current city through authorization, associating it with regional delivery rules, and combining it with the request time period to form features that include environment and time dimensions; or it can be based on extracting unused redemption coupon information from the user's logged-in account, combining it with the device's network status, and combining it into features that include activity and device dimensions.

[0059] Step S203: Input the user profile, the dynamic value coefficient, and the contextual features into the deep model, and output the second feature of the target item.

[0060] It should be noted that the deep model is a deep neural sub-model within the fusion ranking model responsible for uncovering implicit and complex feature associations. It consists of multiple fully connected networks and uses nonlinear transformations to capture high-order interactions between user profiles, dynamic value coefficients, and contextual features. The second feature is a quantitative feature output by the deep model, representing the implicit matching degree between the user and the target item. Its value reflects the degree of fit based on deep feature interactions (such as nonlinear combinations of scenarios, preferences, and values), possessing the ability to capture potential needs and complementing the first feature output by the linear model.

[0061] Understandably, the Deep part (deep neural network model) in the fusion ranking model is a multi-layer feedforward neural network (such as 3 ReLU layers), which is responsible for generalization and can discover the potential and complex nonlinear relationship between user preferences and dynamic value.

[0062] Specifically, the Deep part (deep neural network) is input with user-side features, product-side features, and contextual features. User-side features may include user profile vector U, as well as embedded user ID, region, etc.; product-side features may include dynamic value coefficient V, as well as product category, brand, current popularity of the product, etc.; contextual features C may include current time, whether it is a holiday, etc.

[0063] In one example, the Deep part transforms user profiles and dynamic features into low-dimensional dense vectors through embedding layers and multi-layer neural networks, and learns their high-order interaction relationships, thereby achieving generalization to unseen data. The final output is the prediction score of the deep model, which, combined with the Wide part, generates the final probability prediction for recommendation. The data structure is defined as follows: 1. User Profile Vector: json{ "user_id": "123456789", "basic_features": { "age_bucket": 3, "gender": "male" }, "behavioral_features": { "login_freq_30d": 12.5, "avg_order_value": 450.0 }, "preference_features": { "category_pref": { / / Category preference coefficient "electronics": 0.95, "clothing": 0.23, "food": 0.56 }, "brand_pref": { / / Brand preference coefficient "brand_A": 0.88, "brand_B": 0.67 } }, "points_features": { "total_points": 5840, "points_to_expire_in_30d": 500, "points_acquisition_rate": 200 / / Average daily points earned in the last 30 days }, "last_update_time": "20xx-xx-xx 08:00:00" }

[0064] The user profile vector is defined in JSON format, containing the user's unique identifier "user_id" and features constructed from four core dimensions: basic_features record basic attributes such as user age segmentation and gender; behavioral_features present user behavior habits in numerical form; preference_features quantify the user's preference for different categories and brands through coefficients; points_features cover key data related to user points, while last_update_time ensures the timeliness of the profile; this vector can be generated by flattening the output of a multilayer perceptron (MLP) or by aggregating multi-source features, and is finally transformed into a high-dimensional dense floating-point vector that can be directly used as model input.

[0065] This vector can be formed by the output of a flattened multilayer perceptron (MLP) or by feature aggregation, ultimately resulting in a high-dimensional dense floating-point vector.

[0066] 2. Dynamic Value Coefficient: json{ "product_id": "P10086", "timestamp": "20xx-xx-xx 14:30:05", "cash_price": 299.0, "points_price": 29900, "dynamic_value_coefficient": 1.12, "factors": { "inventory_pressure": 0.85, "promotion_boost": 1.2, "cost_factor": 120.0 } }

[0067] The dynamic value coefficient is defined in JSON format and is used to record the real-time value-related data of a specific target item: product_id is the unique identifier of the item, ensuring that the data is accurately associated with the item; timestamp marks the specific time of the coefficient calculation, reflecting the dynamic attribute; cash_price (item cash price) and points_price (number of points required for redemption) are the basic data for value assessment, used to convert the correspondence between points and cash; the core field dynamic_value_coefficient is the final calculated quantitative indicator, which directly represents the current point redemption cost-effectiveness of the item; the factors object contains the key basis for calculating the coefficient: inventory_pressure (inventory pressure coefficient), promotion_boost (marketing adjustment factor), and cost_factor (cost factor).

[0068] Step S204: Determine the final recommendation score of the target item based on the first feature and the second feature.

[0069] It should be understood that the outputs of the Wide and Deep parts can eventually be combined through a weighted sum (or concatenated and then passed through a linear output layer) to produce the final recommendation score.

[0070] Specifically, the first and second features can be standardized in terms of magnitude. Since the output ranges of linear and deep models may differ, Min-Max normalization should be used to transform both to the same preset range to avoid interference from magnitude differences in the fusion results. Second, the feature fusion strategy should be determined, with dynamic weighted summation preferred, and the initial weights determined through offline training. Next, the weights are substituted to calculate the final recommendation score. If there are special scenarios (such as new users with insufficient explicit behavioral data), the weight of the second feature can be temporarily increased to strengthen the influence of implicit features. Finally, the final recommendation score for each target item is output. The higher the score, the higher the overall fit between the item and the user's needs.

[0071] In this embodiment, a fusion ranking model that includes linear and deep models is adopted. The linear model determines the effective capture of clear business rules and simple feature associations, while the deep model can uncover complex feature interactions and potential patterns, thus balancing the interpretability of the recommendations with the ability to learn complex feature relationships.

[0072] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the exchange recommendation method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0073] This application also provides an exchange recommendation device, please refer to... Figure 6 The redemption recommendation device includes: The feature acquisition module 10 is used to respond to a user recommendation request triggered by a user to acquire the user profile of the user and the dynamic value coefficient of each target item, wherein the target item is a product redeemed using points; The scoring calculation module 20 is used to determine the context features corresponding to the user recommendation request, and input the user profile, the dynamic value coefficient and the context features into the fusion ranking model to output the final recommendation score of the target item; The redemption recommendation module 30 is used to generate a redemption recommendation list based on the final recommendation score.

[0074] The redemption recommendation device provided in this application, employing the redemption recommendation method described in the above embodiments, can solve the technical problem of homogenized recommendation results in existing points redemption recommendation technologies, which reduces user engagement and satisfaction with the points system. Compared with the prior art, the beneficial effects of the redemption recommendation device provided in this application are the same as those of the redemption recommendation method provided in the above embodiments, and other technical features in the redemption recommendation device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0075] This application provides a redemption recommendation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the redemption recommendation method in Embodiment 1 above.

[0076] The following is for reference. Figure 7 The diagram illustrates a structural schematic suitable for implementing the redemption recommendation device in the embodiments of this application. The redemption recommendation device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 7 The redemption recommendation device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0077] like Figure 7 As shown, the redemption recommendation device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the redemption recommendation device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the exchange recommending device to communicate wirelessly or wiredly with other devices to exchange data. Although the figures show exchange recommending devices with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0078] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0079] The redemption recommendation device provided in this application, employing the redemption recommendation method described in the above embodiments, can solve the technical problem of homogenized recommendation results in existing points redemption recommendation technologies, which reduces user engagement and satisfaction with the points system. Compared with the prior art, the beneficial effects of the redemption recommendation device provided in this application are the same as those of the redemption recommendation method provided in the above embodiments, and other technical features in this redemption recommendation device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0080] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0081] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0082] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the exchange recommendation method in the above embodiments.

[0083] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having 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. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0084] The aforementioned computer-readable storage medium may be included in the redemption recommendation device; or it may exist independently and not be assembled into the redemption recommendation device.

[0085] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the exchange recommendation device, cause the exchange recommendation device to perform the exchange recommendation method described above.

[0086] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0087] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0088] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0089] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described redemption recommendation method. This addresses the technical problem of homogenized recommendation results in existing points redemption recommendation technologies, which reduces user engagement and satisfaction with the points system. Compared to existing technologies, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the redemption recommendation method provided in the above embodiments, and will not be elaborated upon here.

[0090] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the redemption recommendation method as described above.

[0091] The computer program product provided in this application can solve the technical problem of homogenized recommendation results in existing points redemption recommendation technologies, which reduces user engagement and satisfaction with the points system. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the redemption recommendation method provided in the above embodiments, and will not be repeated here.

[0092] The above description is only a part of the embodiments of this application and does not limit the scope of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included within the protection scope of this application.

Claims

1. A method for recommending redemptions, characterized in that, The redemption recommendation method includes: In response to a user-triggered user recommendation request, the system obtains the user profile of the user and the dynamic value coefficient of each target item, wherein the target items are goods redeemed using points. Determine the contextual features corresponding to the user recommendation request, and input the user profile, the dynamic value coefficient, and the contextual features into the fusion ranking model to output the final recommendation score of the target item; A redemption recommendation list is generated based on the final recommendation score.

2. The redemption recommendation method as described in claim 1, characterized in that, The step of obtaining the user's user profile and the dynamic value coefficient of each target item in response to a user-triggered user recommendation request includes: Based on the user recommendation request, the user profile corresponding to the user is extracted from the user feature library, and the user profile is a feature vector; Obtain real-time attribute information and score information for each target item, and calculate the dynamic value coefficient of each target item based on the real-time attribute information and the score information.

3. The redemption recommendation method as described in claim 2, characterized in that, Before the step of extracting the user profile corresponding to the user from the user feature library based on the user recommendation request, the method further includes: Collect multi-source heterogeneous data corresponding to the user; The multi-source heterogeneous data is preprocessed and the preprocessed data is converted into features to generate structured feature data, which includes user behavior features, user points features and item attribute features. The user profile corresponding to the user is updated based on the feature data, and the user profile is stored in the user feature database.

4. The redemption recommendation method as described in claim 2, characterized in that, The step of acquiring real-time attribute information and score information of each target item, and calculating the dynamic value coefficient of each target item based on the real-time attribute information and the score information, includes: Obtain real-time attribute information and points information for each target item, and determine calculation factors based on the real-time attribute information. The calculation factors include cost factors, inventory pressure coefficients, and marketing adjustment factors. The value of the goods corresponding to the points information is determined according to the preset exchange ratio. The calculation factor, the integral information, and the commodity value information are input into a preset calculation model to calculate the dynamic value coefficient of each target item.

5. The redemption recommendation method as described in claim 1, characterized in that, The fusion ranking model includes a linear model and a deep model; The step of determining the contextual features corresponding to the user recommendation request, inputting the user profile, the dynamic value coefficient, and the contextual features into the fusion ranking model, and outputting the final recommendation score of the target item includes: The first feature of each of the target items is determined based on the linear model; Determine the contextual characteristics of the user when the user recommendation request is triggered; The user profile, the dynamic value coefficient, and the contextual features are input into the deep model, and the second feature of the target item is output. The final recommended score for the target item is determined based on the first feature and the second feature.

6. The redemption recommendation method as described in any one of claims 1 to 5, characterized in that, After the step of generating the redemption recommendation list based on the final recommendation score, the method further includes: Collect user interaction data on target items in the redemption recommendation list; The interactive behavior data is labeled to generate training samples, which include positive samples and negative samples. The positive samples are the target items that the user redeemed, and the negative samples are the target items that the user did not redeem or click on. The fusion ranking model is updated using the training samples with the goal of minimizing the cross-entropy loss function.

7. A redemption recommendation device, characterized in that, The device includes: The feature acquisition module is used to respond to a user recommendation request triggered by a user, and to acquire the user profile of the user and the dynamic value coefficient of each target item, wherein the target items are goods redeemed using points; The scoring calculation module is used to determine the context features corresponding to the user recommendation request, and input the user profile, the dynamic value coefficient and the context features into the fusion ranking model to output the final recommendation score of the target item; The redemption recommendation module is used to generate a redemption recommendation list based on the final recommendation score.

8. A redemption and recommendation device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the redemption recommendation method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the exchange recommendation method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the redemption recommendation method as described in any one of claims 1 to 6.

Images