The application discloses a recommendation strategy determination method and device, a computer device, a storage medium and a program product

Abstract

The application relates to a recommendation strategy determination method and device, computer equipment, a storage medium and a program product. The method comprises the following steps: determining at least one recommendation strategy combination for each to-be-recommended product according to each to-be-recommended product of a user account; predicting the recommendation success rate of each recommendation strategy in the recommendation strategy combination, determining the evaluation value of each recommendation strategy evaluation item according to the recommendation success rate, and determining the total evaluation value corresponding to the recommendation strategy combination according to the evaluation value and a weight coefficient; determining a candidate recommendation strategy combination and a target recommendation strategy combination according to the total evaluation value, performing product recommendation on the user account according to the target recommendation strategy combination; obtaining a recommendation result, predicting the prediction recommendation result of each candidate recommendation strategy combination in the case that the recommendation result meets a weight coefficient updating condition, and adjusting the weight coefficient according to the prediction recommendation result and the evaluation value. The method can improve the adaptation degree of the recommendation strategy combination and the user.

Description

Technical Field

[0001] This application relates to the field of big data, and in particular to a method, apparatus, computer device, storage medium, and program product for determining recommendation strategies. Background Technology

[0002] With the advancement of technology, platforms can now recommend products to users. When making product recommendations, they often use pre-defined recommendation strategies. As a result, users' sensitivity to these strategies is crucial, as it determines whether they will choose to use the recommended products.

[0003] Take the example of recommending bike passes to users in the shared mobility sector. When determining the discount rate for bike passes shown to users, the list of bike passes displayed is highly personalized, and each pass has different attributes, such as validity period, number of activations, single-use discount amount, and daily usage limit. Therefore, using the same discount rate for all bike passes is clearly unreasonable. In this case, it is necessary to determine different discount rate strategies for each type of bike pass when recommending it.

[0004] Currently, the shared mobility sector typically uses manual methods to determine recommendation strategies. For example, regarding discount rates for bike passes, since there are various types of bike passes, recommendations to users often involve combinations of multiple bike passes. Each type of bike pass can have its own discount rate strategy. Currently, technical personnel manually select a discount rate strategy for each bike pass, construct a recommendation strategy combination, and then use this strategy combination to recommend the corresponding bike pass combination.

[0005] This method of relying on the subjective judgment of technical personnel to formulate recommendation strategy combinations is not only inefficient in formulating recommendation strategy combinations, but also results in low adaptability between the determined recommendation strategy combinations and users, thus leading to poor recommendation performance. Summary of the Invention

[0006] Therefore, it is necessary to provide a method, apparatus, computer device, storage medium, and program product for determining recommendation strategies to address the aforementioned technical problems.

[0007] Firstly, this application provides a method for determining a recommendation strategy. The method includes:

[0008] Based on each product to be recommended for a user account, at least one combination of recommendation strategies is determined for each of the products to be recommended; the combination of recommendation strategies consists of recommendation strategies for each of the products to be recommended.

[0009] For any of the aforementioned recommendation strategy combinations, the recommendation success rate of each recommendation strategy in the recommendation strategy combination is predicted. Based on the recommendation success rate of each recommendation strategy in the recommendation strategy combination, the evaluation value of the recommendation strategy combination for each recommendation strategy evaluation item is determined. Based on each evaluation value and the weight coefficient of each recommendation strategy evaluation item, the total evaluation value corresponding to the recommendation strategy combination is determined.

[0010] Based on the total evaluation value of each of the aforementioned recommendation strategy combinations, candidate recommendation strategy combinations are determined, and a target recommendation strategy combination is determined from each of the candidate recommendation strategy combinations. Based on the target recommendation strategy combination, product recommendations are made to the user account.

[0011] The recommendation results of the target recommendation strategy combination are obtained. If the recommendation results meet the weight coefficient update conditions, the predicted recommendation results of each candidate recommendation strategy combination are predicted. Based on the predicted recommendation results of each candidate recommendation strategy combination and each evaluation value, the weight coefficients of each recommendation strategy evaluation item are adjusted.

[0012] In one embodiment, the prediction of the recommendation success rate of each recommendation strategy in the recommendation strategy combination includes:

[0013] The initial recommendation success rate of each recommendation strategy in the combination of recommendation strategies is predicted;

[0014] If no target recommendation strategy exists in any of the recommended strategy combinations, the initial recommendation success rate of each recommended strategy is used as the recommendation success rate of that strategy, where the target recommendation strategy is a strategy whose recommendation success rate is greater than a success rate threshold; or...

[0015] If the target recommendation strategy exists in each of the recommended strategy combinations, the initial recommendation success rate of the target recommendation strategy is used as the recommendation success rate of the target recommendation strategy, and the recommendation success rates of the other recommended strategies besides the target recommendation strategy are predicted based on the initial recommendation success rate of the target recommendation strategy.

[0016] In one embodiment, predicting the predicted recommendation result for each candidate recommendation strategy combination when the recommendation result satisfies the weight coefficient update condition includes:

[0017] If the recommendation result is a failure, determine the user account category to which the user account belongs, and determine each reference user account corresponding to the user account from the user account category based on the user characteristics of the user account;

[0018] For any of the candidate recommendation strategy combinations, the similarity between the historical target recommendation strategy combinations and the candidate recommendation strategy combinations for each of the reference user accounts is determined, and the predicted recommendation result of the candidate recommendation strategy combination is predicted based on the similarity and the recommendation results of each of the historical target recommendation strategy combinations.

[0019] In one embodiment, adjusting the weight coefficients of each recommendation strategy evaluation item based on the predicted recommendation results of each candidate recommendation strategy combination and each evaluation value includes:

[0020] For any target candidate recommendation strategy combination whose predicted recommendation result is successful, the recommendation strategy evaluation items are sorted according to the evaluation values ​​of each target candidate recommendation strategy combination from largest to smallest to obtain a recommendation strategy evaluation item queue.

[0021] The weight coefficients of each recommendation strategy evaluation item are adjusted according to their order in the queue of each recommendation strategy evaluation item.

[0022] In one embodiment, determining at least one combination of recommendation strategies for each of the products to be recommended for a user account includes:

[0023] Based on the products to be recommended for each user account and the preset constraints on the combination of recommendation strategies, at least one combination of recommendation strategies for each of the products to be recommended is determined.

[0024] In one embodiment, the recommendation strategy includes resource compensation values, and the preset constraints include that each of the resource compensation values ​​in the combination of recommendation strategies is within a preset resource compensation value range, and the relationship between each of the resource compensation values ​​and the target resource compensation value satisfies a preset condition.

[0025] In one embodiment, the preset constraint condition further includes that the resource acquisition value corresponding to the recommended strategy combination is greater than or equal to a preset resource acquisition value threshold.

[0026] Secondly, this application also provides a recommendation strategy determination apparatus. The apparatus includes:

[0027] The first determining module is configured to determine at least one combination of recommendation strategies for each of the products to be recommended for a user account; the combination of recommendation strategies consists of recommendation strategies for each of the products to be recommended.

[0028] The prediction module is used to predict the recommendation success rate of each recommendation strategy in any of the recommendation strategy combinations, determine the evaluation value of the recommendation strategy combination for each recommendation strategy evaluation item based on the recommendation success rate of each recommendation strategy in the recommendation strategy combination, and determine the total evaluation value corresponding to the recommendation strategy combination based on each evaluation value and the weight coefficient of each recommendation strategy evaluation item.

[0029] The second determining module is used to determine candidate recommendation strategy combinations based on the total evaluation value of each of the recommendation strategy combinations, and to determine a target recommendation strategy combination from each of the candidate recommendation strategy combinations, and to recommend products to the user account based on the target recommendation strategy combination.

[0030] An adjustment module is used to obtain the recommendation results of the target recommendation strategy combination, predict the predicted recommendation results of each candidate recommendation strategy combination when the recommendation results meet the weight coefficient update conditions, and adjust the weight coefficients of each recommendation strategy evaluation item according to the predicted recommendation results of each candidate recommendation strategy combination and each evaluation value.

[0031] In one embodiment, the prediction module is further configured to:

[0032] The initial recommendation success rate of each recommendation strategy in the combination of recommendation strategies is predicted;

[0033] If no target recommendation strategy exists in any of the recommended strategy combinations, the initial recommendation success rate of each recommended strategy is used as the recommendation success rate of that strategy, where the target recommendation strategy is a strategy whose recommendation success rate is greater than a success rate threshold; or...

[0034] If the target recommendation strategy exists in each of the recommended strategy combinations, the initial recommendation success rate of the target recommendation strategy is used as the recommendation success rate of the target recommendation strategy, and the recommendation success rates of the other recommended strategies besides the target recommendation strategy are predicted based on the initial recommendation success rate of the target recommendation strategy.

[0035] In one embodiment, the adjustment module is further configured to:

[0036] If the recommendation result is a failure, determine the user account category to which the user account belongs, and determine each reference user account corresponding to the user account from the user account category based on the user characteristics of the user account;

[0037] For any of the candidate recommendation strategy combinations, the similarity between the historical target recommendation strategy combinations and the candidate recommendation strategy combinations for each of the reference user accounts is determined, and the predicted recommendation result of the candidate recommendation strategy combination is predicted based on the similarity and the recommendation results of each of the historical target recommendation strategy combinations.

[0038] In one embodiment, the adjustment module is further configured to:

[0039] For any target candidate recommendation strategy combination whose predicted recommendation result is successful, the recommendation strategy evaluation items are sorted according to the evaluation values ​​of each target candidate recommendation strategy combination from largest to smallest to obtain a recommendation strategy evaluation item queue.

[0040] The weight coefficients of each recommendation strategy evaluation item are adjusted according to their order in the queue of each recommendation strategy evaluation item.

[0041] In one embodiment, the first determining module is further configured to:

[0042] Based on the products to be recommended for each user account and the preset constraints on the combination of recommendation strategies, at least one combination of recommendation strategies for each of the products to be recommended is determined.

[0043] In one embodiment, the recommendation strategy includes resource compensation values, and the preset constraints include that each of the resource compensation values ​​in the combination of recommendation strategies is within a preset resource compensation value range, and the relationship between each of the resource compensation values ​​and the target resource compensation value satisfies a preset condition.

[0044] In one embodiment, the preset constraint condition further includes that the resource acquisition value corresponding to the recommended strategy combination is greater than or equal to a preset resource acquisition value threshold.

[0045] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement any of the methods described above.

[0046] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements any of the above methods.

[0047] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements any of the above methods.

[0048] The aforementioned recommendation strategy determination method, apparatus, computer equipment, storage medium, and program product determine at least one combination of recommendation strategies for the product to be recommended. Based on the recommendation success rate of each strategy in the combination, a total evaluation value is calculated for the recommendation strategy combination, and a target recommendation strategy combination is determined based on the total evaluation value. If the recommendation result of the target recommendation strategy combination meets the weight coefficient update condition, the weight coefficients are adjusted based on the predicted recommendation results and evaluation values ​​of each candidate recommendation strategy combination to make the next determined target recommendation strategy combination more accurate. Compared to manually determining the recommendation strategy combination, this embodiment can automatically determine the target recommendation strategy combination for a user account based on the predicted recommendation success rate and evaluation value, and recommend products to the user based on the target recommendation strategy combination. This improves the efficiency of recommendation strategy combination formulation. Furthermore, formulating the target recommendation strategy combination based on the recommendation success rate is more objective and can improve the fit between the target recommendation strategy combination and the user, thereby improving the product recommendation effect. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating a method for determining a recommendation strategy in one embodiment;

[0050] Figure 2 This is a flowchart illustrating step 104 in one embodiment;

[0051] Figure 3 This is a flowchart illustrating step 108 in one embodiment;

[0052] Figure 4 This is a flowchart illustrating step 108 in one embodiment;

[0053] Figure 5 This is a structural block diagram of a recommended strategy determination device in one embodiment;

[0054] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0056] In one embodiment, such as Figure 1 As shown, a method for determining a recommendation strategy is provided. This embodiment illustrates the method applied to a server; however, it is understood that the method can also be applied to a terminal, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0057] Step 102: Based on each product to be recommended for a user account, determine at least one combination of recommendation strategies for each product to be recommended; the combination of recommendation strategies consists of recommendation strategies for each product to be recommended.

[0058] In this embodiment, the products to be recommended are one or more pre-determined products that need to be recommended to user accounts. The products to be recommended for each user account can be the same or different.

[0059] For each product to be recommended, at least one recommendation strategy needs to be determined. The recommendation strategy can consist of the product's display position and the display method. The product's display position can include its location on the display interface, such as whether it's displayed in the upper right or lower right corner, or the order in which the products are arranged in a list or switching page. The display method can include the timing of the display, such as whether the product is displayed when the user clicks to enter the product browsing page or after the user successfully logs in, or the product price and discount rate displayed. This application embodiment does not specifically limit the method for determining the recommendation strategy for each product to be recommended. One or more fixed recommendation strategies can be pre-set for each product to be recommended. Alternatively, after setting multiple recommendation strategies, a pre-trained neural network can be used to filter the pre-set strategies based on the user characteristics corresponding to the user account, selecting the recommendation strategy that performs best for the current user account and the current product to be recommended.

[0060] A recommendation strategy combination consists of a recommendation strategy for each product to be recommended. For example, given products A and B, where A corresponds to recommendation strategies a1 and a2, and B corresponds to recommendation strategies b1 and b2, the recommendation strategy combination can be a1 and b2, a2 ​​and b2, or a1 and b1. Each recommendation strategy combination can cover all possible combinations of recommendation strategies for each product to be recommended, or it can be a combination of a subset of recommendation strategies. In one example, after obtaining all possible combinations of recommendation strategies, combinations containing conflicting or unsatisfactory strategies can be filtered out, leaving the remaining combinations as the recommendation strategy combination.

[0061] Step 104: For any combination of recommendation strategies, predict the recommendation success rate of each recommendation strategy in the combination. Based on the recommendation success rate of each recommendation strategy in the combination, determine the evaluation value of the combination for each recommendation strategy evaluation item. Based on each evaluation value and the weight coefficient of each recommendation strategy evaluation item, determine the total evaluation value corresponding to the combination of recommendation strategies.

[0062] In this embodiment, a model can be pre-trained to predict the recommendation success rate of a product when recommended using a recommendation strategy, based on the user characteristics corresponding to the user account. This model then predicts the recommendation success rate of each recommendation strategy in each combination. The recommendation success rate refers to the probability that a user will interact with a product after it has been recommended according to a recommendation strategy. In one example, user accounts can be categorized according to user characteristics. For each user account category, the proportion of users who interact with a product after it has been recommended according to a recommendation strategy is calculated. This proportion represents the recommendation success rate of that user account category for that recommendation strategy and the product. During model training, the recommendation success rate of a user account is predicted based on the recommendation success rates of all user account categories and the weights of each user account category in the training samples. The weights of each user account category are then adjusted based on the actual recommendation results when recommending to the user account, until a well-trained model is obtained.

[0063] The recommendation strategy evaluation item is used to evaluate the merits of a recommendation strategy combination in a certain aspect. The higher the evaluation value of a recommendation strategy combination for a certain evaluation item, the better the recommendation strategy combination is in the aspect represented by that evaluation item. The evaluation value of each recommendation strategy evaluation item is obtained by calculating a probability value based on the recommendation success rate of each recommendation strategy in the recommendation strategy combination, and then normalizing the probability value. For example, the recommendation strategy evaluation item may include the failure rate of the recommendation strategy combination (since the lower the failure rate, the better the recommendation strategy combination, the evaluation value of this item can be calculated by subtracting the probability that all recommended products are not recommended successfully from 1), the success rate of the recommendation strategy combination for a single product (the corresponding evaluation value can be calculated by the recommendation strategy with the highest recommendation success rate), the success rate of the recommendation strategy combination for two products (the corresponding evaluation value can be calculated by the probability that at least two recommended products are recommended successfully), and the overall success rate of the recommendation strategy combination (the corresponding evaluation value can be calculated by the sum of the recommendation success rates of each recommended product), etc. The specific settings can be made by those skilled in the art according to actual needs.

[0064] Each recommendation strategy evaluation item has a weight coefficient, which characterizes the importance of that evaluation item to the specific user account when evaluating the combination of recommendation strategies. Each user account has an independent weight coefficient. The initial weight coefficients for user accounts can be set by those skilled in the art based on experience, and can be adjusted subsequently according to the recommendation performance. By weighting and summing the evaluation values ​​according to their weight coefficients, the total evaluation value corresponding to the combination of recommendation strategies can be obtained.

[0065] Step 106: Based on the total evaluation value of each recommendation strategy combination, determine the candidate recommendation strategy combination, and determine the target recommendation strategy combination from among the candidate recommendation strategy combinations. Then, recommend products to the user account based on the target recommendation strategy combination.

[0066] In this embodiment, multiple candidate recommendation strategy combinations can be selected based on the total evaluation value. Then, a target recommendation strategy combination can be chosen from these combinations for product recommendation. The recommendation strategy combinations can be sorted from highest to lowest total evaluation value, and the top-ranked strategy combination (within a predetermined percentage) can be selected as a candidate recommendation strategy combination. Alternatively, a strategy combination with a predetermined ranking can be selected as a candidate recommendation strategy combination. The target recommendation strategy combination can be the candidate recommendation strategy combination with the highest total evaluation value, the candidate recommendation strategy combination with the highest evaluation value for a specific evaluation item, or the candidate recommendation strategy combination with the smallest variance in the evaluation values ​​for each evaluation item, i.e., a candidate recommendation strategy combination that performs relatively evenly across all evaluation items. The specific choice can be made by those skilled in the art based on actual needs, and this embodiment does not impose specific limitations on this.

[0067] Step 108: Obtain the recommendation results of the target recommendation strategy combination. If the recommendation results meet the weight coefficient update conditions, predict the predicted recommendation results of each candidate recommendation strategy combination. Based on the predicted recommendation results of each candidate recommendation strategy combination and each evaluation value, adjust the weight coefficients of each recommendation strategy evaluation item.

[0068] In this embodiment, after making recommendations based on the target recommendation strategy combination, the recommendation results of the target recommendation strategy combination can be obtained. The recommendation results include success results and failure results. Success results indicate that the user has interacted with at least a preset number of the recommended products (for example, interactions may include purchasing, adding to cart, clicking to view, or adding to favorites). Failure results indicate that the user has not interacted with the preset number of the recommended products.

[0069] If the recommendation results meet the update conditions for the weight coefficients (e.g., multiple unsuccessful product recommendations for the user account), it indicates that the current weight coefficient settings do not reflect the true importance of each evaluation item to the user account. Therefore, the weight coefficients can be adjusted. The recommendation results of other user accounts similar to the user account can be referenced to predict the recommendation results for each candidate recommendation strategy combination. For candidate recommendation strategy combinations where the predicted recommendation results are successful, the weight coefficients of the evaluation items with higher evaluation values ​​in these combinations should be increased for use in the next product recommendation.

[0070] In one example, multiple reference user accounts similar to the user account can be identified based on the user characteristics of the user account. The recommendation results of the candidate recommendation strategy combination are then predicted based on the previously used recommendation strategy combinations for the reference user accounts, the similarity between these recommendation strategy combinations and the candidate recommendation strategy combinations, and the recommendation results of the previous recommendation strategy combinations.

[0071] Since the success rate of recommending a specific product to a user account using a particular recommendation strategy generally remains unchanged, assuming the user's characteristics do not alter, the server can pre-predict the success rates of all combinations of recommendation strategies and products for each user account. When product recommendations are needed for a user account, the pre-determined success rates are used to calculate the overall evaluation value, requiring minimal computational power. When weight coefficients need updating, the recommendation results of each candidate recommendation strategy combination are predicted using a more computationally intensive method, and the weight coefficients are adjusted accordingly, thereby reducing server resource consumption.

[0072] The recommendation strategy determination method provided in this application determines at least one combination of recommendation strategies for the product to be recommended, calculates the total evaluation value of the recommendation strategy combination based on the recommendation success rate of each strategy in the combination, and determines the target recommendation strategy combination based on the total evaluation value. If the recommendation result of the target recommendation strategy combination meets the weight coefficient update condition, the weight coefficient is adjusted according to the predicted recommendation result and evaluation value of each candidate recommendation strategy combination to make the next determined target recommendation strategy combination more accurate. Compared with manually determining the recommendation strategy combination, this application embodiment can automatically determine the target recommendation strategy combination for the user account based on the predicted recommendation success rate and evaluation value, and recommend products to the user based on the target recommendation strategy combination. This can improve the efficiency of recommendation strategy combination formulation, and the formulation method based on the recommendation success rate is more objective, which can improve the fit between the target recommendation strategy combination and the user, thereby improving the product recommendation effect.

[0073] In one embodiment, such as Figure 2As shown, in step 104, the prediction success rate of each recommendation strategy in the recommendation strategy combination is obtained, including:

[0074] Step 202: Predict the initial recommendation success rate of each recommendation strategy in the recommendation strategy combination.

[0075] Step 204: If no target recommendation strategy exists in any combination of recommendation strategies, the initial recommendation success rate of each recommendation strategy is used as the recommendation success rate of that strategy. The target recommendation strategy is a recommendation strategy with a success rate greater than a success rate threshold; or...

[0076] Step 206: If a target recommendation strategy exists in each combination of recommendation strategies, the initial recommendation success rate of the target recommendation strategy is taken as the recommendation success rate of the target recommendation strategy. Based on the initial recommendation success rate of the target recommendation strategy, the recommendation success rates of other recommendation strategies besides the target recommendation strategy are predicted.

[0077] In this embodiment, since a user's choice of which products to interact with are influenced by other products the user already owns, the predicted success rate of the recommendation strategy can be used as the initial success rate. If no recommendation strategy has an initial success rate exceeding a success rate threshold, then the initial success rate is used as the final success rate of the recommendation strategy. If a target recommendation strategy has an initial success rate exceeding a success rate threshold (e.g., 90%, 95%, or other high values), it can be assumed that the user already owns the products corresponding to these target recommendation strategies. The probability of the user interacting with products corresponding to other recommendation strategies while already owning these products can then be re-predicted, and this re-predicted probability is used as the success rate of the other recommendation strategies.

[0078] For example, referring to the methods in the aforementioned embodiments, the proportion of users who interacted with a product to be recommended after recommending it to a user account category according to the recommendation strategy can be counted. Furthermore, the proportion of other products already held by users who interacted with the product to be recommended, and the proportion of other products already held by users who did not interact with the product to be recommended, can be counted. Based on this, the proportion of users in that user account category who interacted with the product to be recommended while holding a certain product can be determined. For example, for a product B, the number of users who already hold product B and interacted with the product to be recommended A, and the number of users who already hold product B but did not interact with the product to be recommended A, can be counted to calculate the proportion of users holding product B who interacted with the product to be recommended A. Then, referring to the methods in the aforementioned embodiments, the model is trained using the user's existing products and the product to be recommended as input, and the recommendation success rate as output, to obtain a model that can predict the probability of a user interacting with the product to be recommended while holding a certain product.

[0079] The recommendation strategy determination method provided in this application predicts the initial recommendation success rate of the recommendation strategy. In the case of a target recommendation strategy whose initial recommendation success rate exceeds the success rate threshold, the success rate of other recommendation strategies is re-predicted based on the premise that the user holds the product to be recommended corresponding to the target recommendation strategy. This can take into account the impact of the user holding certain products to be recommended on the user's interaction with other products to be recommended, thereby improving the accuracy of the recommendation success rate determination.

[0080] In one embodiment, such as Figure 3 As shown, in step 108, if the recommendation results meet the weight coefficient update conditions, the predicted recommendation results for each candidate recommendation strategy combination are predicted, including:

[0081] Step 302: If the recommendation result is a failure, determine the user account category to which the user account belongs, and determine the reference user accounts corresponding to the user account from the user account category based on the user characteristics of the user account.

[0082] Step 304: For any candidate recommendation strategy combination, determine the similarity between the historical target recommendation strategy combination and the candidate recommendation strategy combination for each reference user account, and predict the recommended result of the candidate recommendation strategy combination based on the similarity and the recommendation results of each historical target recommendation strategy combination.

[0083] In this embodiment of the application, if the recommendation result of the target recommendation strategy combination is a failure, it is determined that the recommendation result meets the weight coefficient update condition, and the predicted recommendation result of each candidate recommendation strategy combination is predicted.

[0084] User accounts can be categorized into multiple user account categories based on user characteristics. These characteristics can include basic user traits such as age, location, and login client type, as well as traits related to the product being recommended. For example, if the recommended product is a shared electric bicycle riding card, user characteristics could include riding frequency and duration of each ride. Clustering user accounts based on these characteristics yields multiple user account categories. The user account category to which the user account for the currently recommended product belongs can be determined. Then, based on the user characteristics of this user account and the user characteristics of other user accounts within the category, one or more reference user accounts similar to the current user account can be identified from that category.

[0085] This allows us to retrieve historical target recommendation strategy combinations previously used by each reference user account. For each candidate recommendation strategy combination, we determine the similarity between each historical target recommendation strategy combination and the candidate recommendation strategy combination. This similarity is determined based on the similarity between each recommendation strategy and the product targeted by each recommendation strategy in the historical target recommendation strategy combination, and the similarity between each recommendation strategy and the product targeted by each recommendation strategy in the candidate recommendation strategy combination. For each product targeted by a candidate recommendation strategy combination, we can determine whether there exists a product that is relatively similar to the product targeted by the historical target recommendation strategy combination. If so, we determine the similarity between the recommendation strategy for that product and the recommendation strategy for that relatively similar product. If not, we continue processing the next product in the candidate recommendation strategy combination. After determining the similarity of all products to be recommended in the above way, we can average the similarities to obtain the similarity between the candidate recommendation strategy combination and the historical target recommendation strategy combination. The similarity of the product to be recommended can be calculated based on the similarity between the tags of the product to be recommended, and the similarity between recommendation strategies can be calculated based on the similarity between each strategy contained in the recommendation strategy.

[0086] The process is explained using a cycling pass as the recommended product and a discount rate as the recommendation strategy. Cycling passes come in various types, such as a 3-day pass that allows two rides within 3 days (referred to as a 3-day 2-ride pass), a 5-day pass with 4 rides, and a 30-day pass with 30 rides. Assume the candidate recommendation strategy combination includes a 60% discount for the 3-day pass with 2 rides and a 40% discount for the 30-day pass with 30 rides. The historical target recommendation strategy combination includes a 50% discount for the 5-day pass with 4 rides and a 30% discount for the 30-day pass with 20 rides. When determining the similarity between the candidate recommendation strategy combination and the historical target recommendation strategy combination, we first determine if there is a product similar to the 3-day pass with 2 rides in the historical target recommendation strategy combination. Since there is no product similar to the 3-day pass with 2 rides, we continue to determine if there is a product similar to the 30-day pass with 30 rides. Since the 30-day pass with 20 rides is quite similar to the 30-day pass with 30 rides, we can further determine the similarity between the 40% discount and the 30% discount. This similarity is used as the similarity between the candidate recommendation strategy combination and the historical target recommendation strategy group.

[0087] Based on the similarity between the candidate recommendation strategy combination and each historical target recommendation strategy combination, as well as the recommendation results of each historical target recommendation strategy combination, the predicted recommendation result of the candidate recommendation strategy combination can be predicted. For example, weights can be assigned to each historical target recommendation strategy combination based on similarity, with the score of historical target recommendation strategy combinations that have achieved successful recommendations set to 1, and the score of historical target recommendation strategy combinations that have achieved unsuccessful recommendations set to 0. The weighted average of the scores of each historical target recommendation strategy combination is then calculated, and the predicted recommendation result is determined based on whether the average is closer to 1 or 0. Alternatively, a model for predicting recommendation results based on similarity can be pre-trained, and the similarity between the candidate recommendation strategy combination and each historical target recommendation strategy combination can be input into the model to obtain the predicted recommendation result of the candidate recommendation strategy combination. This application does not specifically limit this approach.

[0088] The recommendation strategy determination method provided in this application predicts the recommended results of candidate recommendation strategy combinations based on historical target recommendation strategy combinations of reference user accounts that are similar to the user account. Then, the weight coefficients of each recommendation strategy evaluation item are adjusted based on the predicted recommendation results. This can improve the accuracy of the weight coefficients, thereby improving the accuracy of determining the total evaluation value and the target recommendation strategy combination.

[0089] In one embodiment, such as Figure 4 As shown, in step 108, based on the predicted recommendation results and evaluation values ​​of each candidate recommendation strategy combination, the weight coefficients of each recommendation strategy evaluation item are adjusted, including:

[0090] Step 402: For any target candidate recommendation strategy combination whose predicted recommendation result is successful, sort the recommendation strategy evaluation items according to the evaluation values ​​of the target candidate recommendation strategy combination from largest to smallest to obtain the recommendation strategy evaluation item queue.

[0091] Step 404: Adjust the weight coefficients of each recommendation strategy evaluation item according to their order in the queue of each recommendation strategy evaluation item.

[0092] In this embodiment, target candidate recommendation strategy combinations that predict successful recommendation results can be selected. For each target candidate recommendation strategy combination, the recommendation strategy evaluation items are sorted from largest to smallest according to their evaluation values ​​for each recommendation strategy evaluation item, resulting in a recommendation strategy evaluation item queue.

[0093] Since a recommendation strategy evaluation item ranks relatively high (i.e., low) across all queues, it indicates that each target candidate recommendation strategy combination scores highly on that evaluation item. Therefore, the weight coefficient of that evaluation item can be appropriately increased to make the recommendation strategy combination with a high score on that item more likely to become the target recommendation strategy combination. Similarly, when a recommendation strategy evaluation item ranks relatively low (i.e., high) across all queues, its weight coefficient can be appropriately decreased. The ranking of each recommendation strategy evaluation item in each queue can be calculated. If there are evaluation items whose rankings in each queue are all below the first ranking threshold, the weight coefficient for that evaluation item should be increased; if there are evaluation items whose rankings in each queue are all above the second ranking threshold, the weight coefficient for that evaluation item should be decreased. No weight coefficient adjustment should be made for other evaluation items.

[0094] The recommendation strategy determination method provided in this application ranks the recommendation strategy evaluation items according to the evaluation value of each target candidate recommendation strategy combination with a predicted recommendation result of success. The weight coefficients are adjusted based on the order of the evaluation items in each recommendation strategy evaluation item queue. This allows for higher weights for evaluation items with higher scores and lower weights for evaluation items with lower scores. This makes it easier to identify target recommendation strategy combinations similar to those with predicted successful recommendation results when using the adjusted weight coefficients, thereby improving the accuracy of target recommendation strategy combination determination.

[0095] In one embodiment, step 102, determining at least one combination of recommendation strategies for each product to be recommended based on each product to be recommended for a user account, includes:

[0096] Based on the products to be recommended for each user account and the preset constraints for the combination of recommendation strategies, at least one combination of recommendation strategies for each product to be recommended is determined.

[0097] In this embodiment, the recommended strategy combination is subject to preset constraints. These constraints may include conditions such as the recommended strategies not conflicting with each other and the sum of the benefits from recommending based on each strategy exceeding a threshold. Based on these constraints, the problem of determining the recommended strategy combination can be modeled as a problem of finding all possible solutions under these constraints. Solving this problem yields all possible recommended strategy combinations.

[0098] In one embodiment, the recommendation strategy includes resource compensation values. Preset constraints include that each resource compensation value in the recommendation strategy combination is within a preset resource compensation value range, and the relationship between each resource compensation value and the target resource compensation value satisfies a preset condition. The resource compensation value refers to how much resource compensation a user receives when interacting with the recommended product. Resources can refer to currency, virtual tokens, or any other valuable item. The target resource compensation value is a pre-determined value for the user account, representing the overall resource compensation target for the user account. When recommending products to users through various recommendation strategies, the resource compensation value of each recommendation strategy needs to be close to the target resource compensation value. For example, the average of the resource compensation values ​​of each recommendation strategy can be equal to the target resource compensation value, or the sum of the differences between the average of the resource compensation values ​​of each recommendation strategy and the target resource compensation value can be less than a preset value. Furthermore, the resource compensation value of each recommendation strategy also needs to be within a preset resource compensation value range to limit the compensation range provided by the platform to the user account.

[0099] In one embodiment, the preset constraint further includes that the resource acquisition value corresponding to the recommendation strategy combination is greater than or equal to a preset resource acquisition value threshold. The resource acquisition value refers to the resources the platform expects to obtain after recommending products to a user's account according to the recommendation strategy. This value can be calculated by multiplying the resource value the platform determines it can obtain if the user interacts with the recommended product, and the recommendation success rate of the recommendation strategy. The sum of the resource acquisition values ​​of each recommendation strategy is the resource acquisition value corresponding to the recommendation strategy combination. This value needs to be greater than or equal to the preset resource acquisition value threshold to ensure that the platform can still obtain sufficient resource value from the recommended product after making recommendations according to the recommendation strategy.

[0100] In one embodiment, if it can be determined that updating the weight coefficient is unnecessary when recommending products to a user account (e.g., the weight coefficient update condition is a preset number of consecutive recommendation failures, and even if the current recommendation fails, the number of consecutive recommendation failures will not reach the preset number), then only the target recommendation strategy combination needs to be determined, not the candidate recommendation strategy combination. In this case, the problem of determining at least one recommendation strategy combination and determining the target recommendation strategy combination based on the total evaluation value of each recommendation strategy combination can be modeled as a constrained optimization problem. The optimization objective is to maximize the total evaluation value, and the constraint condition is that the recommendation strategy for each product to be recommended meets certain conditions. Taking the recommendation strategy as resource compensation value, the constraint condition as follows: the resource compensation value of each product to be recommended is within a preset range, the average resource compensation value is equal to the preset resource compensation value determined in advance for the user account, and after each product to be recommended compensates the user according to the resource compensation value, the platform's resource acquisition value is greater than or equal to the preset resource acquisition value, the problem of determining the optimal resource compensation value for each product to be recommended can be modeled as follows:

[0101]

[0102] Formula (1)

[0103] Here, obj represents the optimization objective, that is, maximizing... The value of . The probability of holding the recommended product p under the resource compensation value q, i.e. the recommendation success rate, can be obtained through neural network training. The value can be either 1 or 0. A value of 1 indicates that the recommended product p uses the resource compensation value q, while a value of 0 indicates that the recommended product p does not use the resource compensation value q. The resource compensation value q is an integer between a and b, and n is the total number of products to be recommended.

[0104] This is a recommendation strategy evaluation term, representing the sum of the holding probabilities of each recommended product under a given combination of recommendation strategies. α is the weighting coefficient of this evaluation term. This is another evaluation item for recommendation strategies, representing the probability that a user account will not interact with any of the recommended products under a given combination of recommendation strategies. β is the weighting coefficient for this evaluation item.

[0105] These are the four constraints that the optimization problem described above needs to satisfy. It is a pre-determined target resource compensation value for user accounts. This means that the average resource compensation value of each product to be recommended must be equal to the target resource compensation value. It is the average resource acquisition value of the platform for the user account category to which the user account belongs. Given a resource compensation value q, if a user account purchases the recommended product p, the resource acquisition value that the platform can obtain is defined as the resource transfer value that the platform can obtain when the user interacts with the recommended product, minus the resource transfer value that the user can obtain when using the recommended product. It can be obtained through neural network training. This refers to the fact that under a certain combination of recommendation strategies, the platform's revenue is greater than the average resource acquisition value. It refers to Take 1 or 0. This means that for the product p to be recommended, there must be one and only one. The value is 1, which means that a resource compensation value q must be determined for the product p to be recommended.

[0106] The above problem can be solved using a continuous programming solver or a Lagrange dual optimization to obtain the optimal combination of target recommendation strategies. Based on the resource compensation value corresponding to each product to be recommended in the target recommendation combination, cycling cards can be recommended to user accounts.

[0107] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0108] Based on the same inventive concept, this application also provides a recommendation strategy determination apparatus for implementing the recommendation strategy determination method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the recommendation strategy determination apparatus provided below can be found in the limitations of the recommendation strategy determination method described above, and will not be repeated here.

[0109] In one embodiment, such as Figure 5 As shown, a recommendation strategy determination device 500 is provided, including: a first determination module 502, a prediction module 504, a second determination module 506, and an adjustment module 508, wherein:

[0110] The first determining module 502 is configured to determine at least one combination of recommendation strategies for each of the products to be recommended for a user account; the combination of recommendation strategies consists of recommendation strategies for each of the products to be recommended.

[0111] The prediction module 504 is used to predict the recommendation success rate of each recommendation strategy in any of the recommendation strategy combinations, determine the evaluation value of the recommendation strategy combination for each recommendation strategy evaluation item based on the recommendation success rate of each recommendation strategy in the recommendation strategy combination, and determine the total evaluation value corresponding to the recommendation strategy combination based on each evaluation value and the weight coefficient of each recommendation strategy evaluation item.

[0112] The second determining module 506 is used to determine candidate recommendation strategy combinations based on the total evaluation value of each of the recommendation strategy combinations, and to determine a target recommendation strategy combination from each of the candidate recommendation strategy combinations, and to recommend products to the user account based on the target recommendation strategy combination.

[0113] The adjustment module 508 is used to obtain the recommendation results of the target recommendation strategy combination, predict the predicted recommendation results of each candidate recommendation strategy combination when the recommendation results meet the weight coefficient update conditions, and adjust the weight coefficients of each recommendation strategy evaluation item according to the predicted recommendation results of each candidate recommendation strategy combination and each evaluation value.

[0114] The recommendation strategy determination device provided in this application determines at least one combination of recommendation strategies for a product to be recommended, calculates the total evaluation value of the recommendation strategy combination based on the recommendation success rate of each strategy in the combination, and determines the target recommendation strategy combination based on the total evaluation value. If the recommendation result of the target recommendation strategy combination meets the weight coefficient update condition, the weight coefficient is adjusted according to the predicted recommendation result and evaluation value of each candidate recommendation strategy combination to make the next determined target recommendation strategy combination more accurate. Compared with manually determining the recommendation strategy combination, this application embodiment can automatically determine the target recommendation strategy combination for user accounts based on the predicted recommendation success rate and evaluation value, and recommend products to users based on the target recommendation strategy combination. This can improve the efficiency of recommendation strategy combination formulation, and the formulation method based on the recommendation success rate is more objective, which can improve the suitability of the target recommendation strategy combination with users, thereby improving the product recommendation effect.

[0115] In one embodiment, the prediction module 504 is further configured to:

[0116] The initial recommendation success rate of each recommendation strategy in the combination of recommendation strategies is predicted;

[0117] If no target recommendation strategy exists in any of the recommended strategy combinations, the initial recommendation success rate of each recommended strategy is used as the recommendation success rate of that strategy, where the target recommendation strategy is a strategy whose recommendation success rate is greater than a success rate threshold; or...

[0118] If the target recommendation strategy exists in each of the recommended strategy combinations, the initial recommendation success rate of the target recommendation strategy is used as the recommendation success rate of the target recommendation strategy, and the recommendation success rates of the other recommended strategies besides the target recommendation strategy are predicted based on the initial recommendation success rate of the target recommendation strategy.

[0119] In one embodiment, the adjustment module 508 is further configured to:

[0120] If the recommendation result is a failure, determine the user account category to which the user account belongs, and determine each reference user account corresponding to the user account from the user account category based on the user characteristics of the user account;

[0121] For any of the candidate recommendation strategy combinations, the similarity between the historical target recommendation strategy combinations and the candidate recommendation strategy combinations for each of the reference user accounts is determined, and the predicted recommendation result of the candidate recommendation strategy combination is predicted based on the similarity and the recommendation results of each of the historical target recommendation strategy combinations.

[0122] In one embodiment, the adjustment module 508 is further configured to:

[0123] For any target candidate recommendation strategy combination whose predicted recommendation result is successful, the recommendation strategy evaluation items are sorted according to the evaluation values ​​of each target candidate recommendation strategy combination from largest to smallest to obtain a recommendation strategy evaluation item queue.

[0124] The weight coefficients of each recommendation strategy evaluation item are adjusted according to their order in the queue of each recommendation strategy evaluation item.

[0125] In one embodiment, the first determining module is further configured to:

[0126] Based on the products to be recommended for each user account and the preset constraints on the combination of recommendation strategies, at least one combination of recommendation strategies for each of the products to be recommended is determined.

[0127] In one embodiment, the recommendation strategy includes resource compensation values, and the preset constraints include that each of the resource compensation values ​​in the combination of recommendation strategies is within a preset resource compensation value range, and the relationship between each of the resource compensation values ​​and the target resource compensation value satisfies a preset condition.

[0128] In one embodiment, the preset constraint condition further includes that the resource acquisition value corresponding to the recommended strategy combination is greater than or equal to a preset resource acquisition value threshold.

[0129] Each module in the above-mentioned device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0130] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a recommendation policy determination method.

[0131] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0132] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0133] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0134] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0135] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0136] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0137] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0138] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for determining a recommendation strategy, characterized in that, The method includes: Based on each product to be recommended for a user account, at least one combination of recommendation strategies is determined; the combination of recommendation strategies consists of multiple recommendation strategies, and each product to be recommended has a corresponding recommendation strategy in the combination of recommendation strategies; the recommendation strategy includes at least the position of the product to be recommended on the display interface and / or the way the product to be recommended is displayed. For any of the aforementioned recommendation strategy combinations, the recommendation success rate of each recommendation strategy in the combination is predicted. Based on the recommendation success rate of each recommendation strategy in the combination, the evaluation value of the recommendation strategy combination for each recommendation strategy evaluation item is determined. Based on each evaluation value and the weight coefficient of each recommendation strategy evaluation item, the total evaluation value corresponding to the recommendation strategy combination is determined. The recommendation success rate is used to characterize the probability that a user will interact with the product to be recommended after the product to be recommended is recommended according to the recommendation strategy. The recommendation strategy evaluation item includes at least one of the following: the failure rate of the recommendation strategy combination, the success rate of the recommendation strategy combination for a single product to be recommended, the success rate of the recommendation strategy combination for two products to be recommended, and the sum of the success rates of each product to be recommended in the recommendation strategy combination. Based on the total evaluation value of each of the aforementioned recommendation strategy combinations, candidate recommendation strategy combinations are determined, and a target recommendation strategy combination is determined from each of the candidate recommendation strategy combinations. Based on the target recommendation strategy combination, product recommendations are made to the user account. The recommendation results of the target recommendation strategy combination are obtained. If the recommendation results meet the weight coefficient update conditions, the predicted recommendation results of each candidate recommendation strategy combination are predicted. Based on the predicted recommendation results of each candidate recommendation strategy combination and each evaluation value, the weight coefficients of each recommendation strategy evaluation item are adjusted.

2. The method according to claim 1, characterized in that, The prediction obtains the recommendation success rate of each recommendation strategy in the recommendation strategy combination, including: The initial recommendation success rate of each recommendation strategy in the combination of recommendation strategies is predicted; If no target recommendation strategy exists in any of the recommended strategy combinations, the initial recommendation success rate of each recommended strategy is used as the recommendation success rate of that strategy, where the target recommendation strategy is a strategy whose recommendation success rate is greater than a success rate threshold; or... If the target recommendation strategy exists in each of the recommended strategy combinations, the initial recommendation success rate of the target recommendation strategy is used as the recommendation success rate of the target recommendation strategy, and the recommendation success rates of the other recommended strategies besides the target recommendation strategy are predicted based on the initial recommendation success rate of the target recommendation strategy.

3. The method according to claim 1, characterized in that, The step of predicting the predicted recommendation result for each candidate recommendation strategy combination when the recommendation result satisfies the weight coefficient update condition includes: If the recommendation result is a failure, determine the user account category to which the user account belongs, and determine each reference user account corresponding to the user account from the user account category based on the user characteristics of the user account; For any of the candidate recommendation strategy combinations, the similarity between the historical target recommendation strategy combinations and the candidate recommendation strategy combinations for each of the reference user accounts is determined, and the predicted recommendation result of the candidate recommendation strategy combination is predicted based on the similarity and the recommendation results of each of the historical target recommendation strategy combinations.

4. The method according to claim 3, characterized in that, The step of adjusting the weight coefficients of each recommendation strategy evaluation item based on the predicted recommendation results and evaluation values ​​of each candidate recommendation strategy combination includes: For any target candidate recommendation strategy combination whose predicted recommendation result is successful, the recommendation strategy evaluation items are sorted according to the evaluation values ​​of each target candidate recommendation strategy combination from largest to smallest to obtain a recommendation strategy evaluation item queue. The weight coefficients of each recommendation strategy evaluation item are adjusted according to their order in the queue of each recommendation strategy evaluation item.

5. The method according to claim 1, characterized in that, The step of determining at least one combination of recommendation strategies for each of the products to be recommended for a user account includes: Based on the products to be recommended for each user account and the preset constraints on the combination of recommendation strategies, at least one combination of recommendation strategies for each of the products to be recommended is determined.

6. The method according to claim 5, characterized in that, The recommendation strategy includes resource compensation values, and the preset constraints include that each resource compensation value in the combination of recommendation strategies is within a preset resource compensation value range, and the relationship between each resource compensation value and the target resource compensation value satisfies the preset conditions.

7. The method according to claim 6, characterized in that, The preset constraints also include the resource acquisition value corresponding to the recommended strategy combination being greater than or equal to a preset resource acquisition value threshold.

8. A device for determining a recommendation strategy, characterized in that, The device includes: The first determining module is configured to determine at least one combination of recommendation strategies for each product to be recommended based on each product to be recommended for a user account; the combination of recommendation strategies consists of multiple recommendation strategies, and each product to be recommended has a corresponding recommendation strategy in the combination of recommendation strategies; the recommendation strategy includes at least the position of the product to be recommended on the display interface and / or the way the product to be recommended is displayed. The prediction module is configured to predict the recommendation success rate of each recommendation strategy in any given recommendation strategy combination, determine the evaluation value of the recommendation strategy combination for each recommendation strategy evaluation item based on the recommendation success rate of each recommendation strategy in the recommendation strategy combination, and determine the total evaluation value corresponding to the recommendation strategy combination based on each evaluation value and the weight coefficient of each recommendation strategy evaluation item; the recommendation success rate is used to characterize the probability that a user will interact with the product to be recommended after the product to be recommended is recommended according to the recommendation strategy; wherein, the recommendation strategy evaluation item includes at least one of the following: the failure rate of the recommendation strategy combination, the success rate of the recommendation strategy combination for a single product to be recommended, the success rate of the recommendation strategy combination for two products to be recommended, and the sum of the success rates of each product to be recommended in the recommendation strategy combination; The second determining module is used to determine candidate recommendation strategy combinations based on the total evaluation value of each of the recommendation strategy combinations, and to determine a target recommendation strategy combination from each of the candidate recommendation strategy combinations, and to recommend products to the user account based on the target recommendation strategy combination. An adjustment module is used to obtain the recommendation results of the target recommendation strategy combination, predict the predicted recommendation results of each candidate recommendation strategy combination when the recommendation results meet the weight coefficient update conditions, and adjust the weight coefficients of each recommendation strategy evaluation item according to the predicted recommendation results of each candidate recommendation strategy combination and each evaluation value.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images