An optimization method, device and equipment for resource recommendation and a storage medium

Abstract

This disclosure provides an optimization method, apparatus, device, and storage medium for resource recommendation, relating to the field of computer technology, particularly to the fields of large-scale models, resource recommendation, and intelligent search. The specific implementation scheme includes: determining optimization prompts; inputting the optimization prompts into a large-scale model, which outputs fusion parameters for the resource recommendation system, the fusion parameters including the weights of multiple indicators of the resources in the resource recommendation system; determining the overall lifecycle of the resource based on the fusion parameters and the multiple indicators of the resource; and determining the optimization conclusion for the fusion parameters based on the overall lifecycle and the lifecycle benchmark. This disclosure can improve the accuracy of resource recommendation.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to the fields of large models, resource recommendation, and intelligent search. Background Technology

[0002] In resource recommendation systems, multi-objective fusion is a core element determining the effectiveness of resource recommendations. Currently, most fusion parameters in multi-objective fusion are optimized using evolutionary learning algorithms such as Evolution Strategies (ES). However, these algorithms are highly dependent on human experience and intervention, leading not only to low efficiency in iterative optimization of fusion parameters but also to a lack of characterization of the long-term value of resources, making it difficult to achieve the global optimum of the resource recommendation system. Summary of the Invention

[0003] This disclosure provides an optimization method, apparatus, device, and storage medium for resource recommendation.

[0004] According to one aspect of this disclosure, an optimization method for resource recommendation is provided, comprising: Identify and optimize suggestion keywords; The optimized suggestion word is input into the large model, which outputs the fusion parameters of the resource recommendation system. These fusion parameters include the weights of multiple indicators of the resources in the resource recommendation system. Based on the fusion parameters and multiple indicators of the resources, the overall lifecycle of the resources is determined; Based on the overall lifecycle and lifecycle benchmark, optimization conclusions are determined for the fusion parameters.

[0005] According to another aspect of this disclosure, an optimization apparatus for resource recommendation is provided, comprising: The prompt word determination module is used to determine optimized prompt words; The parameter output module is used to input the optimized suggestion word into the large model, and the large model outputs the fusion parameters of the resource recommendation system, which include the weights of multiple indicators of the resources in the resource recommendation system. The cycle determination module is used to determine the overall lifecycle of the resource based on the fusion parameters and multiple indicators of the resource. The conclusion determination module is used to determine the optimization conclusion for the fusion parameter based on the overall lifecycle and lifecycle benchmark.

[0006] According to another aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and The memory is communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described in the present disclosure.

[0007] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform any of the methods according to embodiments of this disclosure.

[0008] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements any of the methods according to embodiments of this disclosure.

[0009] This disclosure improves the objectivity of determining fusion parameters by inputting optimized prompts into a large model, which then generates fusion parameters for the resource recommendation system. Based on the fusion parameters and multiple resource indicators, the value evolution of resources can be characterized, thereby improving the accuracy of assessing the overall lifecycle of resources. Furthermore, based on the overall lifecycle and lifecycle benchmarks, optimization conclusions are continuously fed back to achieve fine-tuning of the fusion parameters. The mechanism for optimizing fusion parameters proposed in this disclosure improves the accuracy of resource recommendations.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0011] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a schematic diagram illustrating an application scenario according to an embodiment of this disclosure; Figure 2 This is a flowchart illustrating the implementation of an optimization method for resource recommendation according to an embodiment of the present disclosure; Figure 3 This is a flowchart illustrating an optimization method for resource recommendation according to an embodiment of the present disclosure; Figure 4 This is a schematic diagram of the structure of a resource recommendation optimization device 400 according to an embodiment of the present disclosure; Figure 5 This is a schematic diagram of the structure of a resource recommendation optimization device 500 according to an embodiment of the present disclosure; Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement embodiments of the present disclosure is shown. Detailed Implementation

[0012] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0013] The term "and / or" in this disclosure indicates that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The term "at least one" in this document means any combination of at least two of a plurality of options, such as including at least one of A, B, and C, which can mean including any one or more elements selected from the set of A, B, and C. The terms "first" and "second" in this document refer to and distinguish multiple similar technical terms, and do not imply a specific order or a limitation to only two. For example, "first feature" and "second feature" refer to two types / two features; the first feature can be one or more, and the second feature can also be one or more.

[0014] In resource recommendation systems, multi-objective fusion is a core element determining the effectiveness of resource recommendations. Currently, the fusion parameters in multi-objective fusion are mostly optimized using algorithms such as Elasticsearch (ES) and reinforcement learning. However, these algorithms heavily rely on human experience and intervention, leading to low optimization iteration efficiency. Furthermore, in the process of optimizing fusion parameters, existing technologies often limit the setting of reward objectives to short-term goals such as click-through rate, lacking a deep characterization of the long-term value of resources (such as lifetime (LT)). This makes it difficult for resource recommendation systems to achieve global optimization, thus hindering the improvement of the overall performance of resource recommendation systems.

[0015] In view of the above problems, this disclosure proposes a technical solution that integrates a large-scale model and an evolutionary learning algorithm framework, aiming to achieve the autonomous generation and optimization of fusion parameters. This technical solution constructs a complete closed-loop logic of "generation-evaluation-optimization," enabling continuous optimization without manual intervention. It achieves the integration of a large-scale model and an evolutionary learning algorithm to solve the resource recommendation problem.

[0016] Specifically, this disclosure treats the fusion parameters of a resource recommendation system as an optimizable living organism. By simulating the principles of natural selection and genetics, and combining this with the semantic understanding capabilities of a large model, the fusion parameters are generated and iteratively optimized. The goal is to enable the large model not only to output the fusion parameters but also to optimize the fusion parameters and the large model itself. The entire process involves task definition, fusion parameter generation, automatic evaluation, and continuous optimization. By applying the closed-loop logic of "generation-evaluation-optimization" to the resource recommendation problem, this disclosure enables the fusion parameters to continuously approach the optimal solution.

[0017] Figure 1 This is a schematic diagram illustrating an application scenario according to an embodiment of this disclosure, such as... Figure 1 As shown in the illustration, the application scenario diagram of this disclosure may include, but is not limited to, a parameter generation device 110 and a resource recommendation system 120. The parameter generation device 110 and the resource recommendation system 120 can communicate via any type of wired or wireless network. Specifically, the parameter generation device can generate fusion parameters based on the business objectives of resource recommendation, prior knowledge of each resource, etc., and send the fusion parameters to the resource recommendation system 120. The resource recommendation system determines the recommendation score for each resource based on the fusion parameters, and then recommends each resource to the user based on the recommendation score. The resource recommendation system 120 proposed in this disclosure includes, but is not limited to, electronic devices such as mobile phones, computers, smart voice interaction devices, smart home appliances, in-vehicle terminals, game consoles, e-book readers, multimedia playback devices, and wearable devices. Furthermore, this disclosure does not impose a specific limitation on the number of parameter generation devices 110; for example, the application scenario diagram of this disclosure may include one or more parameter generation devices 110.

[0018] Figure 2 This is a flowchart illustrating the implementation of a resource recommendation optimization method according to an embodiment of the present disclosure, including: S210. Determine the optimized prompt keywords; S220. Input the optimized suggestion word into the large model, and the large model outputs the fusion parameters of the resource recommendation system. The fusion parameters include the weights of multiple indicators of resources in the resource recommendation system. S230. Based on the fusion parameters and multiple indicators of the resources, determine the overall lifecycle of the resources; S240. Based on the overall lifecycle and lifecycle benchmark, determine the optimization conclusions for the fusion parameters.

[0019] In this embodiment, optimized prompts are input into a large model. Leveraging the model's semantic understanding and generation capabilities, the model outputs the fusion parameters required for the resource recommendation system. These fusion parameters include the weights of multiple metrics for each resource in the recommendation system. In one example, the fusion parameters can be used to weight and sum multiple metrics for a resource to determine its recommendation score. Based on these scores, resources are then recommended to the user. It is understood that these fusion parameters can be considered control variables at the decision logic level of the resource recommendation system, rather than model variables at the feature learning level.

[0020] For example, multiple metrics for a resource might include click-through rate (CTR), completion rate, and conversion rate. A fusion parameter can serve as the weights for these metrics (the fusion parameter can contain the weights of each metric, typically presented in matrix form). For instance, the weight of CTR could be set to 0.3, completion rate to 0.2, and conversion rate to 0.5. Then, the multiple metrics are weighted and summed according to the fusion parameter to determine the resource's recommendation score. Based on this score, resources can be ranked to obtain a resource recommendation sequence, which is then used to recommend resources to users.

[0021] In this embodiment of the disclosure, the overall lifecycle of a resource can be considered as the overall value of the resource. For example, the overall lifecycle can be a quantitative representation of resource quality and user feedback. In one example, this disclosure recommends various resources based on fusion parameters and multiple indicators of each resource, and then determines the lifecycle of each resource (which can be called an individual lifecycle) based on the recommendation feedback data of each resource. Furthermore, the overall lifecycle of the resource is determined using the individual lifecycles of each resource. The method for determining the overall lifecycle will be described in detail later.

[0022] In this embodiment of the disclosure, the lifecycle benchmark can be a reference standard used to evaluate the value status of resources. In one example, the process of optimizing the fusion parameters may include multiple rounds. Furthermore, the lifecycle benchmark is not a fixed value; it can be adjusted according to the round in which the fusion parameters are optimized. The optimization conclusion can refer to decision information generated after comparing the overall lifecycle of the resource with the lifecycle benchmark, used to guide the optimization of the fusion parameters. For example, the optimization conclusion for the fusion parameters in the previous round can be applied to the optimization process of the fusion parameters in the current round.

[0023] By employing the above method, this disclosure improves the objectivity of determining the fusion parameters by inputting optimized prompts into a large model, which then generates the fusion parameters for the resource recommendation system. Based on the fusion parameters and multiple resource indicators, the value evolution of resources can be characterized, thereby improving the accuracy of assessing the overall lifecycle of resources. Furthermore, based on the overall lifecycle and lifecycle benchmarks, optimization conclusions can be continuously fed back to achieve fine-tuning of the fusion parameters. The mechanism for optimizing fusion parameters proposed in this disclosure improves the accuracy of resource recommendations.

[0024] Figure 3 This is a schematic flowchart of a resource recommendation optimization method according to an embodiment of the present disclosure.

[0025] This disclosure takes the overall lifecycle of a resource as the optimization objective, and combines at least one of the following: business objectives, prior knowledge, and external tools. It utilizes a prompt word sampler 310 to construct optimized prompt words, leverages the semantic understanding and reasoning capabilities of a large model 320 to generate fusion parameters, and applies these fusion parameters to a resource recommendation system 330, which then recommends resources based on these parameters. Subsequently, an experimental evaluation platform 340 collects recommendation feedback data for each resource, and based on this data, determines the overall lifecycle of at least one resource. Further, an evaluator 350 evaluates the overall lifecycle of at least one resource to obtain an optimization conclusion for the fusion parameters, which is then returned to the prompt word sampler 310. The prompt word sampler 310 can then generate new optimized prompt words based on this conclusion to further optimize the fusion parameters in the next round.

[0026] The following content details how to determine optimized prompt words.

[0027] In some implementations, the process of optimizing the fusion parameters of the resource recommendation system includes multiple optimization rounds; Determine the optimized suggestion keywords, including: In the case of the first optimization round, optimization prompts are determined based on at least one of the following: the business objectives of resource recommendation, prior knowledge of each resource, and external tools. If the optimization round is not the first round, the optimization conclusions are integrated into the optimization prompts of the previous optimization round to obtain the optimization prompts for the current optimization round.

[0028] In the embodiments disclosed herein, such as Figure 3 As shown, the process of optimizing the fusion parameters can employ a multi-round optimization mechanism, comprising multiple optimization rounds executed sequentially. For different optimization rounds, the prompt sampler 310 generates optimized prompts in different ways.

[0029] In one example, when the optimization round is the first round, the prompt word sampler 310 can determine the optimized prompt words based on at least one of the following: the business objectives of resource recommendation, the prior knowledge of each resource, and external tools. In this example, the business objectives of resource recommendation can refer to the business goals expected to be achieved in the application scenario of resource recommendation, such as increasing user dwell time, interaction frequency, and number of visits; the prior knowledge of resources can refer to the static attributes of resources (such as videos, products, articles, etc.), including but not limited to resource metadata (such as category, tags, authors, etc.), historical performance indicators (such as historical click-through rate, completion rate, etc.), and hot / cold attributes (such as whether it is a new resource, whether it is in the decline period, etc.); external tools can refer to software modules, service interfaces, and other tools that the large model 320 can actively call during the optimization of fusion parameters. These external tools can be accessed by the large model 320 on demand during the inference process to obtain real-time external information.

[0030] In another example, when the optimization round is not the first round, the prompt sampler 310 can merge the optimization conclusions generated after the previous optimization round into the optimization prompts used in the previous round, thereby dynamically updating and enhancing the prompt content to obtain the optimization prompts for the current optimization round. In this example, fields for placing optimization conclusions can be pre-set in the optimization prompts. The prompt sampler 310 can place the optimization conclusions from the previous optimization round into the corresponding fields of the optimization prompts from the previous optimization round, thereby merging the optimization conclusions from the previous optimization round with the optimization prompts to obtain the optimization prompts for the current optimization round.

[0031] In this embodiment, the optimization conclusion can be an evaluation of the fusion parameters generated by the large model. For example, if the fusion parameters generated in the current optimization round are better than those generated in the previous optimization round, the optimization conclusion can be positive optimization; if the fusion parameters generated in the current optimization round are not better than those generated in the previous optimization round, the optimization conclusion can be negative optimization. In this embodiment, positive optimization can refer to the fusion parameters generated in the current optimization round showing an improvement in resource recommendation performance compared to the previous optimization round, meaning the fusion parameters generated in the current optimization round can more effectively guide resource recommendation; negative optimization can refer to the fusion parameters generated in the current optimization round showing the same or decreased performance in resource recommendation, meaning the fusion parameters generated in the current optimization round fail to effectively guide resource recommendation.

[0032] In this embodiment of the disclosure, the prompt word sampler 310 can dynamically construct optimized prompt words based on at least one of the following: the business objectives of resource recommendation, the prior knowledge of each resource, external tools, and optimization conclusions, so as to guide the large model to iteratively optimize the fusion parameters.

[0033] By adopting the above approach, this disclosure improves the accuracy of the fusion parameters of the resource recommendation system by introducing a multi-round iterative optimization mechanism. In the first optimization round, optimization prompts can be constructed by combining business objectives, prior knowledge, and at least one of external tools, giving the generation of fusion parameters a clear business orientation and data support. In subsequent optimization rounds, the optimization conclusions from the previous round can be fused with the optimization prompts from the previous round to obtain the optimization prompts for the current round, thereby achieving continuous optimization of the fusion parameters. Furthermore, based on the optimization prompts, this disclosure can leverage the semantic understanding and reasoning capabilities of a large model to continuously optimize the fusion parameters, thereby improving the accuracy of resource recommendations.

[0034] like Figure 3 As shown, after the prompt word sampler 310 determines the optimized prompt word, the optimized prompt word can be input into the large model 320, and the large model 320 outputs fusion parameters. Further, the fusion parameters can be applied to the resource recommendation system 330, which recommends at least one resource based on the fusion parameters and determines the overall lifecycle of the resource.

[0035] The following content details how to determine the overall lifecycle of a resource.

[0036] In some implementations, the overall lifecycle of the resources is determined based on fusion parameters and multiple resource metrics, including: Based on the fusion parameters, multiple indicators of each resource are weighted and summed to determine the recommendation score for each resource. Based on the recommendation scores of each resource, a resource recommendation sequence is determined; Based on the resource recommendation sequence, recommendations are made for each resource; Obtain recommendation feedback data for each resource, and based on the recommendation feedback data, determine the overall lifecycle of at least one resource.

[0037] In this embodiment, each resource is typically associated with multiple metrics, which may include click-through rate, completion rate, playback duration, user dwell time, number of interactions, content quality, etc. In one example, the fusion parameter can be the weights of multiple metrics of the resource, generally presented in the form of a weight vector. The dimension of this weight vector is the same as the dimension of the multiple metrics; in other words, the fusion parameter can reflect the relative importance of each metric of the resource. This disclosure encapsulates the fusion parameters generated by the large model into a dictionary in the form of structured key-value pairs. Here, the key can be a metric, and the value can be the weight corresponding to the metric (i.e., the fusion parameter contains the weight of the metric). Then, through a standardized data transmission or configuration distribution mechanism, the fusion parameter is injected into the resource recommendation system to realize the application of the fusion parameter to the resource recommendation system 330.

[0038] In this example, after applying the fusion parameters to the resource recommendation system 330, the fusion layer of the resource recommendation system 330 can multiply each indicator of a resource with the fusion parameters and then sum them (i.e., weighted summation) to obtain the recommendation score for that resource. For each resource, the resource recommendation system 330 can repeat the weighted summation operation of indicators and fusion parameters to obtain the recommendation score for each resource.

[0039] For example, multiple metrics for a resource can include click-through rate (CTR), completion rate, and playback duration. The fusion parameter can be [0.2, 0.3, 0.5], meaning the weight of CTR is 0.2, the weight of completion rate is 0.3, and the weight of playback duration is 0.5. Based on these multiple metrics and fusion parameters, the resource's recommendation rating could be a CTR score. 0.2+ completion rate rating 0.3+ rating based on playback duration 0.5.

[0040] Furthermore, after obtaining the recommendation scores for each resource, the resource recommendation system 330 can sort the resources from highest to lowest recommendation score to determine a resource recommendation sequence. This resource recommendation sequence reflects the resource recommendation system 330's judgment on resource priority; resources with higher recommendation scores are likely to be considered more aligned with business objectives and should be recommended first, thus appearing earlier in the resource recommendation sequence. In one example, the decision layer of the resource recommendation system 330 can determine the resource recommendation sequence based on the recommendation scores of each resource calculated by the fusion layer.

[0041] In this embodiment of the disclosure, the resource recommendation system 330 can apply the generated resource recommendation sequence to the front-end display or service interface, and push the resource recommendation sequence to the user to achieve the recommendation of various resources.

[0042] Furthermore, this disclosure can obtain recommendation feedback data for each resource, and based on the recommendation feedback data, determine the overall lifecycle of at least one resource.

[0043] By employing the above method, this disclosure generates a recommendation score for each resource by weighted summing of the fusion parameters and resource indicators, and determines the resource recommendation sequence accordingly, thereby achieving orderly and efficient resource recommendation. Simultaneously, by acquiring recommendation feedback data for each resource and characterizing the overall lifecycle of the resource based on the recommendation feedback data, this mechanism provides data support for determining the optimization conclusions of the fusion parameters and the next optimization round, thereby optimizing the fusion parameters in a better direction and improving the accuracy of resource recommendation.

[0044] In some implementations, obtaining recommendation feedback data for each resource includes: Using an experimental evaluation platform, user behavior sequences corresponding to each resource are collected. The user behavior sequences include at least one of resource click information, resource conversion information, and resource interaction information. Based on user behavior sequences, the recommendation feedback data for each resource is determined.

[0045] like Figure 3 As shown, this disclosure allows for detailed observation of recommendation feedback data through the experimental evaluation platform 340. In one example, when a resource is pushed to a user, the experimental evaluation platform 340 can record the user behavior sequence related to that resource in real time, i.e., the complete interaction trajectory generated by the user after contacting the resource, which may include: resource click information (i.e., whether the resource was clicked), resource conversion information (i.e., whether a conversion was completed based on the resource), and resource interaction information (i.e., whether the user performed actions such as liking / favoriting / sharing the resource).

[0046] Furthermore, after obtaining the user behavior sequence, the experimental evaluation platform 340 can aggregate, calculate, and map the user behavior sequence according to preset feedback definition rules to generate structured recommendation feedback data. In one example, the recommendation feedback data may include the distribution volume, interaction duration, and number of times each resource is opened. Here, the distribution volume can refer to the total number of times a resource is successfully displayed to users within a specified time period. In this example, the experimental evaluation platform 340 can traverse the user behavior sequence, filter out events of type "exposure" corresponding to each resource, and count and summarize them according to the resource. The interaction duration can refer to the cumulative duration of effective interaction between users and resources, such as video viewing time, article reading time, etc. In this example, the experimental evaluation platform 340 can obtain the interaction duration between each resource and different users from the user behavior sequence, and then obtain the interaction duration of each resource through statistical calculation. The number of times a resource is opened can refer to the number of times a resource is opened by a user. In this example, the experimental evaluation platform 340 can determine the number of times each resource is opened by different users from the user behavior sequence, and then determine the number of times each resource is opened.

[0047] By employing the above method, this disclosure collects user behavior sequences corresponding to each resource through an experimental evaluation platform, and extracts structured recommendation feedback data based on these sequences, thereby achieving a quantifiable evaluation of resource recommendations. Furthermore, this disclosure improves the completeness and timeliness of the recommendation feedback data by basing it on the actual interaction process between users and resources.

[0048] In some implementations, the overall lifecycle of at least one resource is determined based on recommendation feedback data, including: Based on recommendation feedback data, determine the individual lifecycle of each resource; Based on the individual lifecycles of each resource, determine the overall lifecycle of at least one resource.

[0049] In some implementations, the individual lifecycle of each resource is determined based on recommendation feedback data, including: The individual lifecycle of each resource is determined based on at least one of the following: distribution volume, interaction duration, and number of times the resource is opened, as contained in the recommendation feedback data.

[0050] In one example, this disclosure can independently model the lifecycle of each resource across different dimensions based on recommendation feedback data, thereby obtaining the individual lifecycle of each resource.

[0051] Specifically, for the distribution volume, interaction duration, and number of times opened in the resource recommendation feedback data, this disclosure can independently analyze the trend characteristics of each dimension over time (such as growth rate, decay rate, etc.), and determine the life cycle status of the resource in a fixed dimension or calculate the life cycle score of that fixed dimension based on the preset life cycle stage division rules. Thus, the individual life cycle of the resource in three different dimensions (i.e., distribution volume, interaction duration, and number of times opened) can be obtained, and the individual life cycle of each resource can be determined in these three different dimensions, i.e., the individual life cycle of each resource.

[0052] Furthermore, this disclosure can integrate the individual lifecycles of each resource according to corresponding dimensions to determine the integrated lifecycle of at least one resource in different dimensions, that is, to obtain the integrated lifecycle of at least one resource in three different dimensions. Next, in one example, this disclosure can map the integrated lifecycle of at least one resource in three different dimensions to corresponding scores according to a preset mapping rule. Then, this disclosure can perform a weighted summation of the scores corresponding to the integrated lifecycles in the three different dimensions according to a preset weight coefficient to determine the overall lifecycle of at least one resource.

[0053] For example, the dimensions of an individual lifecycle include distribution volume, interaction duration, and number of times it is opened. Then, for each of these dimensions, the individual lifecycle of each resource is determined. In other words, each resource corresponds to an individual lifecycle across these three dimensions. For individual lifecycles within the same dimension, resources can be integrated to obtain an integrated lifecycle for at least one resource in that dimension. Furthermore, the individual lifecycles of each resource are integrated within each dimension to obtain an integrated lifecycle for at least one resource across three different dimensions. In this example, this disclosure maps the integrated lifecycles under each dimension to corresponding scores. The overall lifecycle of at least one resource is determined by a weighted sum of the scores corresponding to the integrated lifecycles across the three different dimensions. In one example, the overall lifecycle can characterize the quantitative evaluation result of the comprehensive activity status of at least one resource during the recommendation process. This overall lifecycle integrates the evolution trends of the three dimensions—distribution volume, interaction duration, and number of times it is opened—to depict the current lifecycle stage of at least one resource.

[0054] By employing the above method, this disclosure constructs individual lifecycles for each resource across multiple dimensions based on recommendation feedback data. Furthermore, based on these individual lifecycles, it generates an overall lifecycle characterizing the comprehensive evolutionary state of at least one resource, achieving a refined depiction of resource value. The overall lifecycle not only improves the accuracy of state perception for at least one resource but also provides quantifiable decision-making basis for subsequent closed-loop optimization of fusion parameters, thereby enhancing the accuracy of resource recommendations.

[0055] The above describes the calculation method for the overall lifecycle. In one example, this disclosure can use the overall lifecycle of at least one resource calculated by operations managers through a strong correlation algorithm to determine the optimization conclusion of the fusion parameters in the current round, and then optimize the fusion parameters for the next round based on the optimization conclusion.

[0056] In another example, such as Figure 3 As shown, the evaluator 350 uses causal analysis to identify core indicators that are strongly correlated with the overall lifecycle of at least one resource, and determines the overall lifecycle accordingly. Based on this, it determines the optimization conclusion for the fusion parameters in this round, and then optimizes the fusion parameters for the next round based on the optimization conclusion.

[0057] The following content details how the optimization conclusions are determined.

[0058] like Figure 3 As shown, this disclosure can utilize evaluator 350 to determine the optimization conclusion.

[0059] In some implementations, optimization conclusions for the fusion parameters are determined based on the overall lifecycle and the lifecycle baseline, including: Determine the difference between the overall lifecycle and the lifecycle baseline; Based on the difference values, the optimization conclusions for the fusion parameters are determined.

[0060] In this embodiment of the disclosure, evaluator 350 can be used to determine the difference between the overall lifecycle and the lifecycle baseline. In one example, the lifecycle baseline may be a lifecycle reference value for the current round of optimization of the fusion parameters.

[0061] In this embodiment of the disclosure, the evaluator 350 can calculate the numerical deviation between the overall lifecycle and the lifecycle baseline to determine the difference between the two. For example, if the overall lifecycle score is 0.65 and the lifecycle baseline score is 0.85, then the difference between the two is -0.2.

[0062] Furthermore, after determining the difference value, the evaluator 350 can determine the optimization conclusion for the fusion parameters based on the difference value.

[0063] In some implementations, optimization conclusions for the fusion parameters are determined based on the difference values, including: When the difference value indicates that the overall lifecycle is better than the lifecycle benchmark, the optimization conclusion of the fusion parameters is determined to be a positive optimization. When the difference value characterizing the overall lifecycle is not better than the lifecycle benchmark, the optimization conclusion for determining the fusion parameters is a non-positive optimization.

[0064] In this disclosure, if the overall lifecycle of at least one resource exceeds the lifecycle benchmark, then the overall lifecycle can be considered superior to the lifecycle benchmark. In one example, the overall lifecycle score is higher than the lifecycle benchmark score, in which case the difference can be a positive number.

[0065] Furthermore, if the difference value indicates that the overall lifecycle is better than the lifecycle benchmark, the optimization conclusion of the fusion parameters can be determined as a positive optimization. In one example, positive optimization can mean that the fusion parameters effectively promote resource recommendation, that is, the fusion parameters play a positive guiding role in resource recommendation.

[0066] In this embodiment of the disclosure, if the overall lifecycle of at least one resource does not exceed a lifecycle benchmark, then the overall lifecycle can be considered not better than the lifecycle benchmark. In one example, the score of the overall lifecycle is not higher than the score of the lifecycle benchmark, in which case the difference value can be a non-positive number.

[0067] Furthermore, if the difference value, representing the overall lifecycle, is not better than the lifecycle benchmark, the optimization conclusion of the fusion parameters can be determined as a non-positive optimization. In one example, non-positive optimization may mean that the fusion parameters failed to effectively promote resource recommendation, or even had an invalid or negative impact on the resource recommendation results; that is, the fusion parameters did not play a positive guiding role in resource recommendation.

[0068] By employing the above method, this disclosure quantitatively compares the overall lifecycle of at least one resource with a lifecycle benchmark, calculates the difference between the two, and accordingly determines the optimization conclusions for the fusion parameters. This enables an objective and interpretable qualitative assessment of the effects of the fusion parameters, thereby improving the targeting and efficiency of fusion parameter iteration.

[0069] Furthermore, such as Figure 3 As shown, after the evaluator 350 determines the optimization conclusion for the fusion parameters, the prompt word sampler 310 can integrate the optimization conclusion into the optimization prompt word of the previous optimization round to update the optimization prompt word of the next optimization round, and then continue to optimize the fusion parameters.

[0070] The following content details how to determine the lifecycle baseline.

[0071] In some embodiments, the method further includes: Determine the lifecycle baseline to be applied to the optimization rounds.

[0072] In some implementations, determining the lifecycle baseline applied to the optimization rounds includes: When the optimization round is the first round, a lifecycle benchmark applicable to the optimization round is determined based on pre-set data; If the optimization round is not the first round, the lifecycle benchmark applied to the optimization round is determined based on the overall lifecycle in the previous optimization round.

[0073] In this embodiment of the disclosure, when the optimization round for the fusion parameters is the first round, a lifecycle benchmark cannot be constructed based on the application results of the fusion parameters due to the lack of historical optimization feedback for the fusion parameters. In this case, the lifecycle benchmark can be determined by pre-set data. In one example, the pre-set data may include, but is not limited to: (1) historical operation data of the resource recommendation system; (2) the lifecycle set by industry or business objectives; and (3) offline simulation data for at least one resource.

[0074] Based on the pre-set data, the life cycle benchmark is determined, providing initial benchmarking data for the generation and evaluation of the first round of fusion parameters, so that the optimization process of fusion parameters can start from a reasonable starting point.

[0075] In this embodiment of the disclosure, when the optimization round for the fusion parameters is not the first round, the present disclosure can accumulate the recommended performance of at least one resource under the current fusion parameters from the previous optimization round. In this case, the lifecycle benchmark for the current optimization round is dynamically determined based on the overall lifecycle of at least one resource obtained in the previous optimization round. In one example, the present disclosure can directly determine the overall lifecycle of at least one resource from the previous optimization round as the lifecycle benchmark for the current optimization round.

[0076] By employing the above method, in the initial optimization of the fusion parameters, an initial lifecycle baseline is determined based on pre-set data. This allows the present invention to initiate optimization of the fusion parameters with reasonable objectives even in the absence of feedback on their application. In subsequent optimization rounds, the lifecycle baseline for the current optimization round is dynamically adjusted based on the overall lifecycle actually determined in the previous optimization round. This mechanism reduces optimization stagnation or target deviation caused by static lifecycle baselines, enhances the directionality and effectiveness of iterative optimization of fusion parameters, and thus improves the accuracy of resource recommendation.

[0077] In some embodiments, the method further includes: If the optimization round is not the first round, the model parameters of the large model are adjusted based on the optimization conclusions.

[0078] In the embodiments disclosed herein, such as Figure 3 As shown, the large model 320 can typically be general or pre-trained, allowing it to generate fusion parameters in the first round of optimization based on its initial inference capabilities. In subsequent optimization rounds, this disclosure determines the optimization conclusion (i.e., positive or negative optimization) regarding the current fusion parameters through a closed-loop evaluation logic. If the optimization conclusion is only used to construct the optimization prompts for the next round without updating the large model 320 itself, it may still repeatedly generate suboptimal fusion parameters. Therefore, this disclosure introduces a mechanism for adjusting the model parameters of the large model 320 based on the optimization conclusions, thereby achieving continuous updates to the model's capabilities.

[0079] In some implementations, the model parameters of the large model are adjusted based on the optimization results, including: Based on the optimization conclusions, the direction for adjusting the model parameters of the large model is determined; Adjust the model parameters of the large model according to the direction of the adjustment.

[0080] In this embodiment of the disclosure, the process of adjusting the model parameters of the large model 320 based on the optimization conclusions includes two stages: Phase 1: After completing the generation, application, feedback and evaluation of the fusion parameters in one round, the evaluator 350 can output the optimization conclusion for the fusion parameters. The optimization conclusion may also contain fine-grained attribution information (i.e. the reasons that led to the optimization conclusion).

[0081] In one example, after performing semantic understanding and structured analysis on the optimization conclusions, this disclosure identifies systematic biases or derivation capabilities in the large model 320 when generating fusion parameters, thereby determining the direction for adjusting the model parameters of the large model 320.

[0082] Second stage: After clarifying the direction of adjustment, this disclosure can use an appropriate model update technique to adjust the model parameters of the large model 320.

[0083] In one example, model parameter tuning could include: constructing supervised training samples using the "optimization cue word-fusion parameter-optimization conclusion" triples from historical optimization rounds for small-scale supervised fine-tuning (SFT); or employing a parameter fine-tuning method based on low-rank adaptation (LoRA), updating only the trainable modules in the large model 320. Through such operations, the large model 320 can gradually internalize multi-round optimization experience, enabling it to generate better fusion parameters even when faced with similar outputs in subsequent inference.

[0084] Using the above method, this disclosure determines the adjustment direction of the model parameters of the large model based on the optimization conclusion, and makes targeted adjustments to the model parameters accordingly. This mechanism enables the next optimization round for the fusion parameters to not only utilize the optimization prompts improved by the optimization conclusion, but also to internalize the optimization conclusion into the knowledge and preferences of the large model itself, thereby improving the accuracy and stability of the large model in generating fusion parameters.

[0085] In some implementations, adjusting the model parameters of a large model involves multiple iterations; Based on the optimization results, the direction for adjusting the model parameters of the large model is determined, including: If the optimization result of the fusion parameters is positive, the adjustment direction of the model parameters in this iteration is determined to be the same as the adjustment direction in the previous iteration. If the optimization conclusion of the fusion parameters is not a positive optimization, the adjustment direction of the model parameters in this iteration is determined to be opposite to the adjustment direction in the previous iteration.

[0086] In this embodiment, adjusting the model parameters of the large model 320 is a multi-round iterative process closely coupled with the optimization rounds of the fusion parameters. After each optimization round of the fusion parameters is completed, if the current optimization round is not the first round, an operation to adjust the model parameters of the large model can be triggered simultaneously. In other words, the iteration rounds of the model parameters correspond to the optimization rounds of the fusion parameters. For example, after the first optimization round of the fusion parameters ends and an optimization prompt for the current optimization round is generated, after the optimization prompt for the current optimization round is input into the large model 320, the large model 320 can determine the direction of model parameter adjustment based on the optimization conclusions contained in the optimization prompt.

[0087] In one example, if the optimization result is positive, it indicates that the fusion parameters effectively promote the lifecycle performance of resources, meaning that the adjustment direction of the model parameters in the previous iteration was correct. In this case, this disclosure can maintain the adjustment direction unchanged, that is, the adjustment direction of the model parameters in this iteration is consistent with the adjustment direction of the model parameters in the previous iteration, so as to further enhance the ability of the large model 320 to generate fusion parameters.

[0088] In another example, if the optimization result is a non-positive optimization, it indicates that the current fusion parameters have not achieved the expected effect, meaning that the adjustment direction of the model parameters in the previous iteration may have introduced a deviation. In this case, this disclosure can reverse the adjustment direction, that is, the adjustment direction of the model parameters in the current iteration is opposite to the adjustment direction of the model parameters in the previous iteration, in order to correct the adjustment direction of the model parameters.

[0089] By employing the above method, this disclosure designs the adjustment of model parameters for a large model as a multi-round iterative process associated with the optimization rounds of the fusion parameters, and dynamically adjusts the direction of model parameter adjustment based on the optimization conclusions. Specifically, when the optimization conclusion of the fusion parameters is positive, the adjustment direction of the previous iteration is continued to strengthen the model parameters that have been verified to be effective; when the optimization conclusion of the fusion parameters is negative, the adjustment direction is reversed to promptly correct the deviation of the large model and reduce error accumulation. In this way, the accuracy and stability of adjusting model parameters can be improved.

[0090] In some embodiments, the method further includes: During the M iterations, if the optimization conclusions of the fusion parameters are all non-positive optimizations, the model parameters are regenerated, where M is greater than or equal to a pre-set threshold.

[0091] In this embodiment, determining the direction for adjusting the model parameters of the large model based on the optimization conclusions includes not only continuing or reversing the adjustment direction based on the adjustment results of a single round of model parameters, but also a model parameter reset mechanism: during M iterations, if the optimization conclusions of the fused parameters are all non-positive optimizations, it can be determined that the model parameter adjustment strategy for the large model has failed. In this case, the original adjustment direction can no longer be continued or reversed, and the model parameters of the large model can be regenerated. It is understood that M can be an integer greater than or equal to a preset threshold. For example, if the preset threshold is 10, then M is an integer greater than or equal to 10.

[0092] In one example, the process of regenerating model parameters involves random seeding. Specifically, random seeding does not involve a complete initialization of the entire model; instead, it involves injecting small-amplitude Gaussian noise into the trainable modules of the large model, or generating several sets of candidate model parameters in parallel based on multiple different random seeds. These sets of candidate model parameters can be used with the same optimization cue words to generate fused parameters. Furthermore, the effectiveness of the fused parameters generated from each set of model parameters can be quickly evaluated in offline simulations or low-volume experiments, thereby selecting model parameters with positive potential.

[0093] By employing the above method, this disclosure proactively triggers a model parameter regeneration mechanism if, during the M-round iteration process, the optimization conclusion of the fusion parameters is detected to be consistently non-positive. This reduces the performance degradation of large models. Furthermore, this model parameter regeneration mechanism breaks the limitations of the original adjustment direction by introducing random point distribution, enabling large models to restart the exploration of high-value fusion parameters.

[0094] This disclosure also proposes an optimization apparatus for resource recommendation. Figure 4 This is a schematic diagram of the structure of a resource recommendation optimization device 400 according to an embodiment of the present disclosure, comprising: The prompt word determination module 410 is used to determine optimized prompt words; The parameter output module 420 is used to input the optimized suggestion word into the large model, and the large model outputs the fusion parameters of the resource recommendation system, which include the weights of multiple indicators of the resources in the resource recommendation system. The cycle determination module 430 is used to determine the overall life cycle of the resource based on the fusion parameters and multiple indicators of the resource. The conclusion determination module 440 is used to determine the optimization conclusion for the fusion parameter based on the overall life cycle and the life cycle benchmark.

[0095] In some implementations, the period determination module 430 is used for: Based on the fusion parameters, multiple indicators of each resource are weighted and summed to determine the recommendation score for each resource. Based on the recommendation scores of each resource, a resource recommendation sequence is determined; Based on the resource recommendation sequence, recommendations are made for each resource; Obtain recommendation feedback data for each resource, and based on this recommendation feedback data, determine the overall lifecycle of at least one resource.

[0096] In some implementations, the period determination module 430 is used for: Based on recommendation feedback data, determine the individual lifecycle of each resource; Based on the individual lifecycles of each resource, determine the overall lifecycle of at least one resource.

[0097] In some implementations, the period determination module 430 is used for: The individual lifecycle of each resource is determined based on at least one of the following: distribution volume, interaction duration, and number of times the resource is opened, as contained in the recommendation feedback data.

[0098] In some implementations, the period determination module 430 is used for: Using an experimental evaluation platform, collect user behavior sequences corresponding to each resource. These user behavior sequences include at least one of resource click information, resource conversion information, and resource interaction information. Based on user behavior sequences, the recommendation feedback data for each resource is determined.

[0099] In some implementations, the conclusion determination module 440 is used for: Determine the difference between the overall lifecycle and the lifecycle baseline; Based on this difference value, the optimization conclusions for the fusion parameters are determined.

[0100] In some implementations, the conclusion determination module 440 is used for: When the difference value indicates that the overall lifecycle is better than the lifecycle benchmark, the optimization conclusion of the fusion parameters is determined to be a positive optimization. When the difference value characterizing the overall lifecycle is not better than the lifecycle benchmark, the optimization conclusion for determining the fusion parameters is a non-positive optimization.

[0101] In some implementations, the process of optimizing the fusion parameters of the resource recommendation system includes multiple optimization rounds; Prompt word determination module 410 is used for: In the case of the first optimization round, optimization prompts are determined based on at least one of the following: the business objectives of resource recommendation, prior knowledge of each resource, and external tools. If the optimization round is not the first round, the optimization conclusions are integrated into the optimization prompts of the previous optimization round to obtain the optimization prompts for the current optimization round.

[0102] In some embodiments, this disclosure also proposes an optimization apparatus for resource recommendation. Figure 5 This is a schematic diagram of the structure of a resource recommendation optimization device 500 according to an embodiment of the present disclosure, comprising: The baseline determination module 550 is used to determine the lifecycle baseline applied to the optimization round.

[0103] In some implementations, the reference determination module 550 is used for: When the optimization round is the first round, a lifecycle benchmark applicable to the optimization round is determined based on pre-set data; If the optimization round is not the first round, the lifecycle benchmark applied to the optimization round is determined based on the overall lifecycle in the previous optimization round.

[0104] In some implementations, the resource recommendation optimization apparatus further includes a model adjustment module 560, used for: If the optimization round is not the first round, the model parameters of the large model are adjusted based on the optimization conclusions.

[0105] In some implementations, the model adjustment module 560 is used for: Based on the optimization conclusions, the direction for adjusting the model parameters of the large model is determined; Adjust the model parameters of the large model according to the direction of the adjustment.

[0106] In some implementations, adjusting the model parameters of a large model involves multiple iterations; Model adjustment module 560 is used for: If the optimization result of the fusion parameters is positive, the adjustment direction of the model parameters in this iteration is determined to be the same as the adjustment direction in the previous iteration. If the optimization conclusion of the fusion parameters is not a positive optimization, the adjustment direction of the model parameters in this iteration is determined to be opposite to the adjustment direction in the previous iteration.

[0107] In some implementations, the model adjustment module 560 is also used for: During the M iterations, if the optimization results of the fusion parameters are all non-positive optimizations, the model parameters are regenerated, and M is greater than or equal to a pre-set threshold.

[0108] The specific functions and examples of each module and submodule of the apparatus in this disclosure can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.

[0109] The acquisition, storage, and application of personal information by users involved in this technical solution comply with relevant laws and regulations and do not violate public order and good morals.

[0110] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0111] Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0112] like Figure 6 As shown, device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 602 or a computer program loaded into random access memory (RAM) 603 from storage unit 608. RAM 603 may also store various programs and data required for the operation of device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0113] Multiple components in device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of monitors, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows device 600 to exchange / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0114] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as resource recommendation optimization methods. For example, in some embodiments, the resource recommendation optimization methods may be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of the resource recommendation optimization methods described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to perform resource recommendation optimization methods by any other suitable means (e.g., by means of firmware).

[0115] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0116] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0117] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0118] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0119] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0120] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0121] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0122] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An optimization method for resource recommendation, comprising: Identify and optimize suggestion keywords; The optimized prompt words are input into a large model, and the large model outputs the fusion parameters of the resource recommendation system. The fusion parameters include the weights of multiple indicators of resources in the resource recommendation system. Based on the fusion parameters and multiple indicators of the resource, the overall lifecycle of the resource is determined; Based on the overall lifecycle and lifecycle benchmark, optimization conclusions are determined for the fusion parameters.

2. The method according to claim 1, wherein, The determination of the overall lifecycle of the resource based on the fusion parameters and multiple indicators of the resource includes: Based on the fusion parameters, multiple indicators of each resource are weighted and summed to determine the recommendation score of each resource. Based on the recommendation scores of each resource, a resource recommendation sequence is determined; Based on the resource recommendation sequence, each of the resources is recommended. Obtain recommendation feedback data for each of the resources, and determine the overall lifecycle of at least one of the resources based on the recommendation feedback data.

3. The method according to claim 2, wherein, Determining the overall lifecycle of at least one of the resources based on the recommendation feedback data includes: Based on the recommendation feedback data, the individual lifecycle of each of the resources is determined; Based on the individual lifecycles of each of the resources, determine the overall lifecycle of at least one of the resources.

4. The method according to claim 3, wherein, Determining the individual lifecycle of each resource based on the recommendation feedback data includes: The individual lifecycle of each resource is determined based on at least one of the distribution volume, interaction duration, and number of times the resource is opened, as contained in the recommendation feedback data.

5. The method according to claim 2, wherein, The step of obtaining recommendation feedback data for each of the resources includes: Using an experimental evaluation platform, user behavior sequences corresponding to each of the resources are collected, and the user behavior sequences include at least one of resource click information, resource conversion information, and resource interaction information. Based on the user behavior sequence, recommendation feedback data for each of the resources is determined.

6. The method according to any one of claims 1-5, wherein, The step of determining the optimization conclusions for the fusion parameters based on the overall lifecycle and lifecycle benchmark includes: Determine the difference between the overall lifecycle and the lifecycle baseline; Based on the difference value, an optimization conclusion is determined for the fusion parameters.

7. The method according to claim 6, wherein, The step of determining the optimization conclusion for the fusion parameters based on the difference value includes: If the difference value indicates that the overall lifecycle is better than the lifecycle benchmark, the optimization conclusion of the fusion parameter is determined to be a positive optimization. If the difference value indicates that the overall lifecycle is not better than the lifecycle benchmark, the optimization conclusion of the fusion parameter is determined to be a non-positive optimization.

8. The method according to any one of claims 1-7, wherein the process of optimizing the fusion parameters of the resource recommendation system includes multiple optimization rounds; The determination of optimized prompt words includes: In the case that the optimization round is the first round, the optimization prompt words are determined based on at least one of the following: the business objective of resource recommendation, the prior knowledge of each of the resources, and external tools. If the optimization round is not the first round, the optimization conclusion is integrated into the optimization prompt words of the previous optimization round to obtain the optimization prompt words of the current optimization round.

9. The method according to claim 8, further comprising: Determine the lifecycle baseline to be applied to the optimization round.

10. The method according to claim 9, wherein, The determination of the lifecycle benchmark applied to the optimization round includes: When the optimization round is the first round, the life cycle benchmark applied to the optimization round is determined based on pre-set data; If the optimization round is not the first round, the life cycle benchmark applied to the optimization round is determined based on the overall life cycle in the previous optimization round.

11. The method of claim 8, further comprising: If the optimization round is not the first round, the model parameters of the large model are adjusted based on the optimization conclusion.

12. The method according to claim 11, wherein, The adjustment of the model parameters of the large model based on the optimization conclusion includes: Based on the optimization conclusions, the direction for adjusting the model parameters of the large model is determined; Based on the direction of the adjustment, the model parameters of the large model are adjusted.

13. The method according to claim 12, wherein, The adjustment of the model parameters of the large model includes multiple iterations; The step of determining the direction of adjusting the model parameters of the large model based on the optimization conclusion includes: If the optimization conclusion of the fusion parameters is positive optimization, the adjustment direction of the model parameters in this iteration is determined to be the same as the adjustment direction in the previous iteration. If the optimization conclusion of the fusion parameters is not a positive optimization, the adjustment direction of the model parameters in this iteration is determined to be opposite to the adjustment direction in the previous iteration.

14. The method of claim 12, further comprising: During the M iterations, if the optimization conclusions of the fusion parameters are all non-positive optimizations, the model parameters are regenerated, and M is greater than or equal to a pre-set threshold.

15. An optimization apparatus for resource recommendation, comprising: The prompt word determination module is used to determine optimized prompt words; The parameter output module is used to input the optimized suggestion words into the large model, and the large model outputs the fusion parameters of the resource recommendation system. The fusion parameters include the weights of multiple indicators of resources in the resource recommendation system. The cycle determination module is used to determine the overall lifecycle of the resource based on the fusion parameters and multiple indicators of the resource; The conclusion determination module is used to determine the optimization conclusion for the fusion parameters based on the overall lifecycle and the lifecycle benchmark.

16. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-14.

17. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-14.

18. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-14.

Images