Methods, devices, equipment, and storage media for obtaining service configuration parameters

Abstract

This disclosure provides a method, apparatus, electronic device, storage medium, and computer program product for obtaining business configuration parameters, relating to the field of artificial intelligence technology, particularly to the fields of artificial intelligence, machine learning, intelligent search, and intelligent recommendation technology. The specific implementation scheme is as follows: acquiring traffic request data; inputting the feature data of the traffic request data into at least one set of target models to obtain the output parameters of each set of target models; the output parameters of each set of target models represent the business indicator gain resulting from changing the business processing of the traffic request data from using the original business configuration parameters corresponding to each set of target models to using the candidate business configuration parameters corresponding to each set of target models; based on the business indicator gain of each set of target models, determining the target business configuration parameters from the original business configuration parameters and candidate business configuration parameters corresponding to each set of target models. According to this disclosure, accurate acquisition of target business configuration parameters can be achieved.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, particularly to the fields of artificial intelligence, machine learning, intelligent search, and intelligent recommendation, and more specifically to methods, devices, electronic devices, storage media, and computer program products for obtaining business configuration parameters. Background Technology

[0002] Research on massive data services such as search, advertising, and product recommendations has revealed that the selection of service configuration parameters can significantly impact service revenue and user experience. Therefore, providing accurate service configuration parameters has become a pressing technical challenge. Summary of the Invention

[0003] This disclosure provides a method, apparatus, storage medium, and computer program product for obtaining business configuration parameters.

[0004] According to one aspect of this disclosure, a method for obtaining business configuration parameters is provided, including:

[0005] Acquire traffic request data; input the feature data of the traffic request data into at least one set of target models to obtain the output parameters of each set of target models; wherein, the output parameters of each set of target models represent the business indicator gain generated when the traffic request data is processed by changing from the original business configuration parameters corresponding to each set of target models to the candidate business configuration parameters corresponding to each set of target models; based on the business indicator gain of each set of target models, determine the target business configuration parameters from the original business configuration parameters and candidate business configuration parameters corresponding to each set of target models, and the target business configuration parameters are used to perform business processing on the object that generates the traffic request data.

[0006] According to another aspect of this disclosure, an apparatus for obtaining service configuration parameters is provided, comprising:

[0007] The first acquisition unit is used to acquire traffic request data;

[0008] The second acquisition unit is used to input the feature data of the traffic request data into at least one set of target models to obtain the output parameters of each set of target models; wherein, the output parameters of each set of target models represent the business indicator gain generated when the traffic request data is processed by changing from the original business configuration parameters corresponding to each set of target models to the candidate business configuration parameters corresponding to each set of target models.

[0009] The determining unit is used to determine the target business configuration parameters from the original business configuration parameters and candidate business configuration parameters corresponding to each group of target models based on the business indicator gain of each group of target models. The target business configuration parameters are used to perform business processing for the object that generates the traffic request data.

[0010] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to said at least one processor; wherein,

[0011] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the methods in any embodiment of this disclosure.

[0012] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing a computer to perform the methods of any embodiment of this disclosure.

[0013] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the methods of any embodiment of this disclosure.

[0014] According to this disclosure, the target business configuration parameters are obtained based on the business indicator gain quantity output by the target model. This takes into account the influence of business indicators on the determination or selection of business configuration parameters, and can achieve accurate acquisition of target business configuration parameters.

[0015] 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

[0016] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0017] Figure 1 This is a schematic diagram of the system architecture of the massive data service system according to an embodiment of this disclosure;

[0018] Figure 2 This is a flowchart illustrating the method for obtaining service configuration parameters according to an embodiment of this disclosure. Figure 1 ;

[0019] Figure 3 This is a flowchart illustrating the method for obtaining service configuration parameters according to an embodiment of this disclosure. Figure 2 ;

[0020] Figure 4This is a schematic diagram of the implementation components of the method for obtaining service configuration parameters according to an embodiment of this disclosure;

[0021] Figure 5 This is a schematic diagram of a method for obtaining service configuration parameters for an implementation component according to an embodiment of this disclosure;

[0022] Figure 6 This is a schematic diagram of a method for obtaining service configuration parameters for an online decision-making component according to an embodiment of this disclosure;

[0023] Figure 7 This is a schematic diagram of the composition of the device for obtaining service configuration parameters according to an embodiment of this disclosure;

[0024] Figure 8 This is a block diagram of an electronic device used to implement the method for obtaining service configuration parameters according to embodiments of the present disclosure. Detailed Implementation

[0025] 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 and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0026] In massive data services, such as search, advertising, and product recommendation, any automatically configurable parameter can be considered a business configuration parameter. For example, the number of product recommendations provided to the user or the number of products displayed on the user's screen each time can be considered business configuration parameters.

[0027] Figure 1 This is the system architecture of a massive data service system according to an embodiment of this disclosure. For example... Figure 1 As shown, this system architecture typically includes a front-end 11 and a back-end 12. The front-end 11 is any reasonable terminal used to display search results, advertising recommendations, or product recommendations provided to the user by the back-end 12. The terminal can be at least one of a mobile phone, all-in-one computer, tablet computer, wearable smart device, etc. The back-end 12 is a server used to provide search results, advertising results, or product recommendations to the front-end 11 using a funnel-shaped architecture, enabling the front-end 11 to display these results to the user.

[0028] The funnel architecture uses a layered approach to filter massive amounts of data based on user-input search keywords or product keywords. It goes through at least several stages, including data retrieval, coarse data ranking, and fine data ranking, before returning the final search or recommendation results to the user.

[0029] Taking product recommendation as an example, the recall phase involves placing products matching the user's input keywords from a massive dataset into a candidate dataset based on the user's request. The coarse ranking phase roughly sorts the candidate data in the candidate dataset. The fine ranking phase, combining user characteristics such as user preferences and purchasing habits, refines the ranking results obtained in the coarse ranking phase to better match user needs.

[0030] Each of the above stages involves specific business configuration parameters. For example, the candidate dataset length is required in the recall stage; the data recommendation length (sorting length for coarse sorting) is required in the coarse sorting stage; and the data recommendation length (sorting length for fine sorting) is required in the fine sorting stage. It is evident that precise configuration of business parameters for each stage will inevitably lead to more accurate product recommendations, resulting in a better user experience. From the system's perspective, it can provide users with more precise business configuration parameters, significantly enhancing the system's business capabilities, such as recommendation capabilities. Such a system is easy to promote in engineering and has strong practicality.

[0031] The method for obtaining business configuration parameters disclosed herein can be used to determine or select business configuration parameters at any of the aforementioned stages. Since this disclosure considers the impact of business metrics on the determination or selection of business configuration parameters, it employs a model (target model) that yields the gain of business metrics to obtain these gains. Based on the gain of business metrics obtained from the target model, the target business configuration parameters are accurately acquired. This scheme for obtaining business configuration parameters based on the gain of business metrics not only effectively improves the user's business experience but also brings revenue to the business. For details, please refer to the subsequent detailed explanation of this scheme.

[0032] The processing logic of the method for obtaining business configuration parameters in this disclosure can be deployed on any suitable server. The server includes a general-purpose server, a cloud server, and servers used in specialized fields as described above. Figure 1 The server in the middle can be used as the backend 12. The device for obtaining service configuration parameters in this embodiment can be used as an implementation component in the server, or it can be implemented independently as a standalone device to separately implement the processing logic of the method for obtaining service configuration parameters of this disclosure.

[0033] The method for obtaining service configuration parameters according to embodiments of this disclosure will be described below. Figure 2 As shown, the method includes:

[0034] S201: Obtain traffic request data;

[0035] Traffic request data can be user search requests, recommendation requests, etc. Users can input search requests and recommendation requests through the front end 11. In this step, the back end 12 can obtain traffic request data by receiving search requests and recommendation requests sent by the front end 11.

[0036] Traffic request data can be from a single user or from multiple users. If it's a single user request, the subsequent solution yields the target service configuration parameters determined for that single user. If it's a request from multiple users, the subsequent solution yields the target service configuration parameters determined for each user. These target service configuration parameters can be completely different, partially the same, or completely identical, depending on the specific circumstances.

[0037] S202: Input the feature data of the traffic request data into at least one set of target models to obtain the output parameters of each set of target models; wherein the output parameters of each set of target models represent the business indicator gain generated when the traffic request data is processed by changing from the original business configuration parameters corresponding to each set of target models to the candidate business configuration parameters corresponding to each set of target models.

[0038] The characteristic data of traffic request data includes, but is not limited to, the following: user attributes, such as gender, age, network age, etc.; user device attributes, such as device model, location data, etc.; user behavior attributes, such as access behavior in a recent period, such as the number of times the user accessed the backend in the last 7 days, the number of times the user interacted with the product merchant in the last day, etc.

[0039] In this step, each target model corresponds to two business configuration parameters: original business configuration parameters and candidate business configuration parameters. Original business configuration parameters are the business configuration parameters originally used by the backend 12 during business processing, such as the original recommendation length provided by the backend 12 in recommendation or search businesses. Candidate business configuration parameters are the business configuration parameters that may be adopted during business processing, such as the candidate recommendation length provided by the backend 12 in recommendation or search businesses.

[0040] Each target model in this embodiment can perform two functions. The first function is to process traffic request data using its corresponding original service configuration parameters. The second function is to process traffic request data using its corresponding candidate service configuration parameters. Each target model generates an output parameter for the aforementioned two functions. This output parameter represents the service metric gain resulting from changing the processing of traffic request data from using the original service configuration parameters corresponding to the target model to using the candidate service configuration parameters corresponding to the target model. That is, the output parameter of each target model represents the service metric gain.

[0041] In application scenarios involving massive amounts of data, such as search, advertising, and recommendation, business metrics can be indicators of business focus, such as click-through rate, purchase rate, satisfaction rate, read rate, and conversion rate. The original recommendation length and candidate recommendation length can specifically be the optional length of the candidate dataset in the recall phase of the aforementioned application scenarios, the optional data recommendation length in the coarse ranking phase, or the optional data recommendation length in the fine ranking phase. From the user's perspective, users can see products presented using either the original recommendation length or the candidate recommendation length.

[0042] In practical applications, the gain can be positive or negative. For internet companies or organizations, a positive gain is ideal, as it indicates business revenue, and the larger the positive gain, the higher the revenue.

[0043] The original service configuration parameters and candidate service configuration parameters in this embodiment can be selected from pre-set parameters. Taking the recommended data length for the fine-ranking stage as an example, the pre-set recommended data length includes values ​​of 100, 200, 300, 400, 500, etc., and one of these can be selected as the original service configuration parameter, and at least one of these can be selected as the candidate service configuration parameter. For example, a recommended data length of 100 is selected as the original service configuration parameter, and recommended data lengths of 200 and 300 are selected as candidate service configuration parameters.

[0044] Since each target model group corresponds to one candidate service configuration parameter, the number of target model groups can be consistent with the number of candidate service configuration parameters. Selecting several parameters as candidate service configuration parameters results in the corresponding number of target model groups, and the traffic request data is then input into these corresponding number of target model groups.

[0045] In practical applications, the target model is any reasonable model that can calculate the gain of business indicators, such as a statistical model that calculates the gain based on statistics, or a gain model based on causal inference.

[0046] S203: Based on the business indicator gain of each group of target models, determine the target business configuration parameters from the original business configuration parameters and candidate business configuration parameters corresponding to each group of target models. The target business configuration parameters are used to perform business processing for the objects that generate traffic request data.

[0047] In this step, based on the output parameters of each group of target models, target business configuration parameters are selected from the original business configuration parameters and candidate business configuration parameters corresponding to each group of target models. The object that generates traffic request data is the user who generates the traffic request data.

[0048] In the aforementioned S201-S203, the target business configuration parameters are obtained based on the business indicator gain output by the target model. Considering the impact of business indicators on the determination or selection of business configuration parameters, a model (target model) that yields the business indicator gain is used to obtain the business indicator gain, thereby achieving accurate acquisition of the target business configuration parameters. This scheme of obtaining business configuration parameters based on business indicator gain not only helps users determine accurate business configuration parameters and improves the user's business experience, but also brings benefits to the business by obtaining the target business configuration parameters based on the business indicator gain, demonstrating the ease of use and practicality of the method in this embodiment.

[0049] As an alternative implementation, such as Figure 3 As shown, the aforementioned scheme for determining the target business configuration parameters from the original business configuration parameters and candidate business configuration parameters corresponding to each target model, based on the business indicator gain of each target model, can be implemented through the following:

[0050] S303: Determine the target business indicator gain from the business indicator gain of each group of target models; wherein, the target business indicator gain is at least one of the business indicator gain values ​​of each group of target models that is the largest and that reaches the target threshold.

[0051] S304: Based on the gain of the target business indicators, determine the target business configuration parameters from the original business configuration parameters and candidate business configuration parameters corresponding to each group of target models.

[0052] exist Figure 3 In the above, the schemes shown in S301 and S302 can be understood by referring to the schemes shown in S201 and S202.

[0053] In S303 to S304, the following situations exist:

[0054] If the target business indicator gain is the largest among the business indicator gain values ​​output by each group of target models, then the candidate business configuration parameter corresponding to the target model with the largest business indicator gain value among the business indicator gain values ​​output by each group of target models will be used as the target business configuration parameter.

[0055] For example, assume there are two candidate service configuration parameters, parameter 1 and parameter 2. The first set of target models corresponds to parameter 1, and the second set of target models corresponds to parameter 2. Among the first and second sets of target models, the first set of target models outputs the largest service indicator gain, so parameter 1 is selected as the target service configuration parameter.

[0056] If the target business indicator gain is the business indicator gain that reaches the target threshold among the business indicator gain outputs of each group of target models, then the candidate business configuration parameters corresponding to the target models that have reached the target threshold among the business indicator gain outputs of each group of target models will be used as the target business configuration parameters.

[0057] For example, assume there are three candidate service configuration parameters: parameter 1, parameter 2, and parameter 3. The first group of target models corresponds to parameter 1, the second group of target models corresponds to parameter 2, and the third group of target models corresponds to parameter 3. Among the first to third group of target models, if the service indicator gain output by the third group of target models reaches the target threshold, while the service indicator gain output by the second and first group of target models does not reach the target threshold, then parameter 3 will be used as the target service configuration parameter.

[0058] If the target business indicator gain is the largest among the business indicator gain values ​​output by each group of target models and reaches the target threshold, then the candidate business configuration parameters corresponding to the target model with the largest business indicator gain value that reaches the target threshold will be obtained from the business indicator gain values ​​output by each group of target models.

[0059] For example, assume there are three candidate service configuration parameters: parameter 1, parameter 2, and parameter 3. The first group of target models corresponds to parameter 1, the second group of target models corresponds to parameter 2, and the third group of target models corresponds to parameter 3. Among the first to third groups of target models, the second group of target models outputs the largest service indicator gain and reaches the target threshold; therefore, parameter 2 is selected as the target service configuration parameter.

[0060] It can be seen that, in Figure 3 In the illustrated scheme, the target business indicator gain is determined based on the business indicator gain that maximizes the value and / or reaches the target threshold. Then, the candidate business configuration parameters corresponding to the target model with the largest business indicator gain, and / or the candidate business configuration parameters corresponding to the target model with the business indicator gain reaching the target threshold, are used as the target business configuration parameters. This achieves accurate acquisition of the target business configuration parameters while maximizing business benefits. It is easy to implement and highly usable in engineering.

[0061] In practical applications, the original business configuration parameters corresponding to each group of target models can be the same or different. As an optional implementation, the original business configuration parameters corresponding to each group of target models are the same. Based on this, the output parameters of each group of target models represent the business indicator gain resulting from changing the business processing of traffic request data from using the same original business configuration parameters to using the candidate business configuration parameters corresponding to each group of target models. If the business indicator gain output by each group of target models does not reach the target threshold, in response to the fact that the business indicator gain output by each group of target models does not reach the target threshold, the aforementioned same original business configuration parameters are used as the target business configuration parameters.

[0062] In layman's terms, if the business metric gain output by each target model fails to reach the target threshold, the original business configuration parameters are used as the target business configuration parameters. This achieves accurate acquisition of the target business configuration parameters, enabling users to determine precise business configuration parameters and effectively improving the user experience.

[0063] In the aforementioned scheme, the target threshold can be a value set based on experience, such as 0.5 or 0.7. Furthermore, the target threshold can also be calculated using a set of target models. For example, the target threshold is obtained by selecting a threshold from at least one preset candidate threshold based on historical business indicator gains obtained from historical traffic request data using at least one set of target models, computing power consumption under historical business configuration parameters corresponding to the historical business indicator gains, and preset business computing power constraints. Further, the computing power consumption can be constrained under business computing power constraints, and the target threshold can be selected from at least one preset candidate threshold to achieve the calculation of the target threshold. Essentially, the target threshold is a value obtained based on computing power resources. It can be understood that the calculation of the target threshold is to achieve more accurate acquisition of business configuration parameters. Therefore, the aforementioned calculation of the target threshold based on computing power consumption and business computing power constraints, and the acquisition of business configuration parameters based on the target threshold, can be considered a scheme to achieve accurate acquisition of business configuration parameters under the action or constraint of business computing power constraints. On the one hand, it can effectively save computing power; on the other hand, it does not affect the normal operation of the business.

[0064] Among them, the business computing power constraint is a condition related to the computing power resources provided by the backend 12, reflecting the computing power provided by the backend 12 to the business. It can be any reasonable condition. For example, the backend 12 originally provides 100 processors to maintain computing power consumption. For every increase in the value of the business configuration parameters, such as 100 or 150, the backend 12 needs to add certain computing resources, such as adding 100 or 150 processors, to support the computing power consumption caused by the increase in business configuration parameters.

[0065] In this embodiment, the impact of the computing power and computing power consumption provided by the backend 12 on the business configuration parameters is considered. The computing power consumption is constrained under the business computing power constraint, and the business configuration parameters are obtained based on the business indicator gain. Under the business computing power constraint, the business effect can be maximized. For example, the candidate configuration parameters corresponding to the target model that achieves the largest business indicator gain are selected as the target business configuration parameters, where maximizing the business indicator gain reflects maximizing the business effect.

[0066] Considering that in application scenarios involving massive amounts of data such as search, advertising, and recommendation, business metrics can be indicators of business focus, such as click-through rate, purchase rate, satisfaction rate, read rate, and conversion rate. Therefore, the candidate business configuration parameters in this embodiment can be the candidate recommendation length in the search or recommendation business, and the original business configuration parameters can be the original recommendation length in the search or recommendation business. Based on this, the business metric gain can be the business benefit generated when each group of target models changes from using the original recommendation length corresponding to each group of target models for business search or business recommendation to using the candidate recommendation length corresponding to each group of target models for business search or business recommendation of traffic request data.

[0067] In the aforementioned solution, the business recommendation length in application scenarios such as search, advertising, and recommendation can be used as a business configuration parameter. The business benefits generated by changing the target model's response to traffic request data from using the original recommendation length corresponding to the target model for business search or recommendation to using the candidate recommendation length corresponding to the target model can be used as a business metric gain. This solution is highly adaptable to the aforementioned application scenarios and has excellent practicality.

[0068] In this context, business revenue can refer to changes in key business metrics, such as increases in click-through rate or purchase rate. It can also refer to the economic benefits derived from these metrics, such as the economic benefits from increases in click-through rate or purchase rate.

[0069] In application scenarios involving massive data services such as search, advertising, and recommendation, the aforementioned business metric gain based on each group of target models, determined from the original business configuration parameters and candidate business configuration parameters corresponding to each group of target models, can be as follows: Based on the business benefits of each group of target models, the target recommendation length is determined from the original recommendation length and candidate recommendation length corresponding to each group of target models. The target recommendation length is used to perform business search or business recommendation for objects that generate traffic request data, such as users.

[0070] Based on business revenue from application scenarios such as search, advertising, and recommendation, the acquisition of target recommendation length can not only ensure the accuracy of recommendation length acquisition but also maximize revenue and improve economic value.

[0071] As an alternative approach, each target model in this embodiment can be a pre-trained gain model based on causal inference, obtained by training each gain model based on causal inference with historical traffic request data and tag data.

[0072] The tag data is obtained based on feedback data of the business configuration parameters recommended to the object by the object in the past 12 historical traffic request data. For example, the product recommendation lengths recommended to the object by the 12 historical backends were 100 or 50. The click data of the object when the product recommendation length was 100 or when the product recommendation length was 50 was collected. Based on the click data, the click-through rate (CTR) of the object when the product recommendation length was 100 or when the product recommendation length was 50 was calculated, and the CTR was used as the tag data. The historical traffic request data and the CTR were used as training samples to train a gain model based on causal inference (referred to as the gain model or uplift model).

[0073] By training the gain model using historical traffic request data and tag data, an accurate target model can be obtained. This accurate target model allows for precise measurement of business metric gain, thereby ensuring the accurate acquisition of target business configuration parameters.

[0074] The following is combined with Figures 4-6 The technical solutions of the embodiments of this disclosure will be described in detail.

[0075] like Figure 4 As shown, the components implementing the method for obtaining service configuration parameters according to embodiments of this disclosure include an online exploration component, an offline training component, and an online decision-making component. The online exploration component is used to acquire historical traffic request data. The offline training component is used to train at least one set of Uplift models using the historical traffic request data and tag data as training samples. The online decision-making component is used to determine target service configuration parameters for the user who generated the traffic request data using the online traffic request data, the trained Uplift model, and the target threshold provided by the offline training component.

[0076] The following example uses the data recommendation length in the fine-ranking stage of a recommendation business application scenario as a business configuration parameter, combining the above components and... Figure 5 The method of the present disclosure embodiment will be described in detail below.

[0077] It's understandable that, based on historical traffic request data, such as historical recommendation requests, backend 12 might recommend different recommendation lengths to different users. For example, the recommendation length provided to user 1 might be 100, to user 2 200, to user 3 300, and so on. These different lengths can be considered as different tiers of business configuration parameters during the fine-tuning phase. Each tier corresponds to a recommendation length. For example, tier 1 corresponds to a recommendation length of 100; tier 2 corresponds to a recommendation length of 200.

[0078] The backend 12 can collect historical traffic request data from users at different tiers. The online exploration component collects or extracts the same proportion of historical traffic request data from users at least one tier. For example, it collects or extracts 1% of the historical traffic request data from users at tier 1 (recommended length 100), 1% from users at tier 2 (recommended length 200), and 1% from users at tier 3 (recommended length 300).

[0079] Among the recommended lengths for different gear levels, one gear level was selected as the control group, and the recommended lengths for the other gear levels were selected as the experimental group.

[0080] For example, a recommended length of 100 can be selected as the control group and used as the original business configuration parameter (original recommended length), while other recommended length values ​​can be selected as the experimental group and used as the candidate business configuration parameter (candidate recommended length).

[0081] The number of uplift models to be trained and the number of candidate service configuration parameters are kept consistent.

[0082] For example, with an original recommendation length of 100 and candidate recommendation lengths of 200 and 300, there are two sets of uplift models to be trained. The first set of uplift models has a candidate recommendation length of 200, and the second set has a candidate recommendation length of 300.

[0083] Both sets of Uplift models to be trained can switch from using the original recommended length to using their respective candidate recommended lengths when processing business requests. For example, the first set of Uplift models to be trained implements: processing historical traffic request data using a recommended length of 100 (the original recommended length); and processing historical traffic request data using a recommended length of 200 (the candidate recommended length). The second set of Uplift models to be trained processes historical traffic request data using a recommended length of 100 (the original recommended length); and processing historical traffic request data using a recommended length of 300 (the candidate recommended length).

[0084] It is understandable that the online exploration component sends the collected or extracted historical traffic request data to the offline training component, which then extracts the feature data from the historical traffic request data.

[0085] The offline training component collects user feedback behavioral data. Business metrics are calculated based on this behavioral data. This behavioral data refers to the actual click-through rate (CTR) or purchase rate of products recommended to users over a historical period, given a specific recommendation length provided by the backend. The offline training component uses the received historical traffic request data and the calculated tag data as training samples to train each set of Uplift models. For example, training the two sets of Uplift models mentioned above.

[0086] Taking the training of the first set of Uplift models as an example, since the original recommendation length of this Uplift model is 100 and the candidate recommendation length is 200, the feature data of the historical traffic request data input to this Uplift model includes the feature data of historical traffic request data of tier 1 and tier 2. Its label data can be the user's actual click-through rate when the recommendation length is 100 and the user's actual click-through rate when the recommendation length is 200.

[0087] The training process is roughly as follows: The feature data from the two historical traffic request data sets are input into the Uplift model to be trained. The Uplift model then outputs a probability value. This probability value represents the degree to which business metrics such as click-through rate (CTR) or purchase rate will change when the Uplift model processes the business from the original recommendation length of 100 to the candidate recommendation length of 200. For example, a probability value of 30% indicates that the business metrics such as CTR or purchase rate have changed by 30%, or increased by 30%, when processing the business from the original recommendation length of 100 to the candidate recommendation length of 200. Given the actual click-through rate of users with a recommendation length of 100 and the probability value output by the Uplift model to be trained, the predicted value of business metrics such as CTR can be obtained when the candidate recommendation length is 200. For example, given that the actual click-through rate (CTR) is 20% when the recommendation length is 100, and the probability value output by the Uplift model to be trained is 30%, then the predicted value of the business metric when the candidate recommendation length is 200 is (1 + probability value) * actual CTR = (1 + 30%) * 20% = 26%. That is, changing the product recommendation from a length of 100 to a length of 200 will result in a 26% CTR, an increase compared to the original length of 100.

[0088] Based on the predicted click-through rate (CTR) and the actual CTR of users with a recommendation length of 200, the loss function is calculated. Training is completed when the loss function is minimized, i.e., the predicted CTR closely matches the actual CTR of users with a recommendation length of 200. The loss function can be any reasonable function, such as the mean squared error function or the squared error function.

[0089] In this application scenario, the uplift model to be trained can be a Multi-Treatment X-Learner algorithm model. It can be understood that the feature data input to the uplift model includes user attributes such as gender and age, device attributes such as device model and location, and user behavior attributes such as the number of visits in a recent period. If each attribute is considered as an input feature of the uplift model to be trained, the Multi-Treatment X-Learner algorithm model has the ability to autonomously filter features based on these input features. That is, it has the ability to filter out features that are useless and useful for model training.

[0090] For example, regarding the input feature of gender, if analysis reveals that for male users, the probability value obtained by the Uplift model to switch from the original recommendation length of 100 to a candidate recommendation length of 200 is 0.3; for female users, the probability value is 0.28. From the perspective of gender, the probability values ​​of 0.3 and 0.28 are not significantly different, indicating that using gender as a basis for predicting business metrics is not very meaningful. Gender is considered a useless input feature and can be disregarded in subsequent iterations of the Uplift model. Regarding the input feature of region, if analysis reveals that for Shanghai users, the probability value obtained by the Uplift model to switch from the original recommendation length of 100 to a candidate recommendation length of 200 is 0.7. For Beijing users, the probability value obtained by changing the recommendation length from 100 to 200 for business processing in the Uplift model to be trained is 0.3. From the perspective of region, the probability values ​​of 0.3 and 0.28 are quite different. As a useful input feature, region should be given priority consideration in the subsequent iterative training of the Uplift model.

[0091] The uplift model to be trained has the above characteristics, which can minimize the amount of computation during model iteration training or adjustment, and shorten the training time.

[0092] It is understandable that the training process for the uplift models to be trained in other groups can be found in the aforementioned explanation of the first group, and the repetitions will not be repeated.

[0093] In this application scenario, after the Uplift models to be trained are completed, the target models for each group are obtained. The trained target models can be sent from the offline training component to the online decision-making component for use, or sent to the online decision-making component for use after passing the evaluation.

[0094] In this application scenario, the following evaluation method can be used to evaluate the trained target model. The offline training component collects test samples, and the model is evaluated based on these samples. Taking target model 1 (which changes from using the original recommendation length of 100 to using a candidate recommendation length of 200 for business processing) as an example, target model 1 can be simply referred to as the target model that changes from Tier 1 (corresponding to a recommendation length of 100) to Tier 2 (corresponding to a recommendation length of 200). Its test samples include traffic request data for Tier 1 and Tier 2. Compared to historical traffic request data, the traffic request data in the test samples can be the current traffic request data. Feature data from the traffic request data of the two tiers is extracted and input into target model 1. The higher the positive probability value output by target model 1, the greater the improvement in business metrics after changing from Tier 1 to Tier 2, and the higher the economic benefits for internet companies. The greater the improvement in business metrics, such as reaching the preset improvement threshold, the better the target model has been trained, and the model is considered to have passed the evaluation.

[0095] The above training and evaluation processes can both be completed by offline training components.

[0096] By training the Uplift model to be trained using historical traffic request data and label data, an accurate target model can be obtained. Based on the model evaluation method, the accuracy of the target model is further guaranteed.

[0097] After each target model is trained or after training is completed and passed evaluation, the offline training component can perform offline simulation of computing power consumption based on historical traffic data, so as to determine the target threshold from at least two preset candidate thresholds under the constraints of business computing power.

[0098] Specifically, the simulation algorithm is as follows:

[0099] A. Based on experience, pre-configure or set two or more candidate thresholds.

[0100] For example, configure three candidate thresholds with values ​​of 0, 0.5 and 0.8 respectively.

[0101] B. Select one of the three candidate thresholds as the score threshold for the uplift model;

[0102] For example, 0 can be chosen as the score threshold for the model.

[0103] C. Input the feature data of historical traffic request data into each target model, and each target model outputs its own probability value.

[0104] For example, the feature data of the historical traffic request data of level 1 is input into each group of target models;

[0105] Assume that Target Model 1 in each group of target models represents a change from processing business data using the original recommended length (the recommended length of Tier 1, with a value of 100) to processing business data using the candidate recommended length of Tier 2 (with a value of 200). Target Model 2 in each group of target models represents a change from processing business data using the original recommended length to processing business data using the candidate recommended length of Tier 3 (with a value of 300). Inputting the feature data of historical traffic request data of Tier 1 into Target Model 1, the output parameter of Target Model 1 represents the business metric gain, such as the gain in click-through rate (CTR), obtained by changing from processing business data using the original recommended length to processing business data using the candidate recommended length of Tier 2 (Gain 1). Inputting the feature data of historical traffic request data of Tier 1 into Target Model 2, the output parameter of Target Model 2 represents the business metric gain, such as the gain in click-through rate (CTR), obtained by changing from processing business data using the original recommended length to processing business data using the candidate recommended length of Tier 3 (Gain 2). Here, Gain 1 and Gain 2 are used as historical business metric gains.

[0106] In all target models, determine which target models have business metric gains exceeding the uplift model's score threshold; among all target models, the candidate recommendation length corresponding to the target model that exceeds the uplift model's score threshold is used as the new recommendation length for users who generate historical traffic request data for Tier 1.

[0107] For example, assuming there are two target models, target model 1 and target model 2, where the gain 1 of target model 1 does not exceed the score threshold of the uplift model, and the gain 2 of target model 2 exceeds the score threshold of the uplift model, then the candidate recommendation length (300) corresponding to target model 2 is used as the new recommendation length for the user who generated the historical traffic request data. The original recommendation length for user 1 who generated the historical traffic request data is 100, and the new recommendation length is 300. Subsequently, the new recommendation length of 300 can be used to make business recommendations for user 1.

[0108] If both gain 1 and gain 2 exceed the score threshold of the uplift model, the candidate recommendation length corresponding to the target model with the largest gain is selected as the new recommendation length for users of historical traffic request data in tier 1. For example, if gain 2 is greater than gain 1, the candidate recommendation length (300) corresponding to target model 2 is selected as the new recommendation length for users of historical traffic request data in tier 1.

[0109] If neither gain 1 nor gain 2 exceeds the score threshold of the uplift model, then the original recommended length of 100 is maintained.

[0110] In this way, the new recommended length can be recalculated for all users who generate historical traffic request data within a certain (historical) period, or the user can be kept at the original recommended length, based on the actual size of the gain.

[0111] In practical applications, the new recommended length can be larger or smaller than the original recommended length. For example, if the original recommended length is 100, the new recommended length can be 50 or 200. The new recommended length can be used as a historical business configuration parameter corresponding to the historical business metric gain.

[0112] D. Obtain the average recommendation length of all users who generated historical traffic request data during this historical period, and convert the average recommendation length into the computing power consumption during this historical period.

[0113] Here, computing power consumption refers to the computing power resources consumed by the backend 12 to provide computing power support for all users during this historical period.

[0114] Assuming there are 100 users in this historical period, and the original recommendation length is 100, after step C, the new recommendation length is calculated to be 50 for 50 users and 150 for the other 50 users. Therefore, the average recommendation length = the sum of the new recommendation lengths for all users / the total number of users = (50*50 + 50*150) / 100 = 100.

[0115] Based on experience, an average recommendation length of 100 requires 100 processors from backend 12 to meet the computational power consumption. An average recommendation length of 200 requires 200 processors from backend 12 to meet the computational power consumption. Therefore, the computational power consumption for an average recommendation length of 100 is approximately equal to or equal to the computational power consumption of 100 processors.

[0116] E. Based on the computing power consumption in step D, determine whether a reasonable candidate threshold needs to be selected from other candidate thresholds as the final score threshold of the uplift model, i.e., the target threshold, according to the business computing power constraints. Specifically, select one candidate threshold from the other candidate thresholds as the score threshold of the uplift model, and repeat steps C, D, and E until the business computing power constraints are met and a suitable candidate threshold is found.

[0117] The business computing power constraint is that the backend 12 can originally provide 100 processors to maintain computing power consumption. For every increase in the recommended length value, such as 50, the backend 12 needs to add certain computing resources, such as 50 processors, to support the computing power consumption caused by the increase in business configuration parameters.

[0118] Compared to historical traffic request data within the same historical period, if the average recommendation length calculated for each user during this historical period, when converted to computing power consumption, can be met by the backend without increasing processor usage, then the uplift model's score threshold of 0.5 is appropriate. If increasing processor usage is required to meet the computing power consumption, then the uplift model's score threshold of 0.5 is not the most suitable score threshold. In this case, steps C, D, and E need to be followed to determine if there is a most suitable score threshold among the other candidate thresholds.

[0119] Compared to historical traffic request data within the same historical period, when the uplift model score threshold is 0.8, the backend can meet the computing power consumption without increasing the processor. When the uplift model score threshold is 0.5, an additional processor is required to meet the computing power consumption. This indicates that the uplift model score threshold of 0.8 is appropriate.

[0120] As can be seen, the business computing power constraint conditions set in this disclosure can select a threshold that can effectively reduce computing power consumption from multiple candidate thresholds as the target threshold. Taking into account the impact of computing power consumption and business computing power constraints on the selection of the target threshold makes the selection of the target threshold more resource-efficient.

[0121] After the offline training component completes the offline computing power simulation, the optimal uplift score threshold (i.e., the target threshold) and the trained target models are synchronized to the online decision component. Each target model can be recorded in the online decision component as a file.

[0122] The online decision-making component loads the target model files and, in the presence of user traffic request data, combines... Figure 6 As shown, the target recommendation length for this user is determined using the following method.

[0123] 1. Extract the feature data of the user's traffic request data and input it into each group of target models;

[0124] 2. Calculate and output the business benefits, such as click-through rate gain, generated by the target model of each group when processing the user from the original recommendation length (e.g., 100) to the candidate recommendation length corresponding to each group's recommendation model;

[0125] 3. From the business revenue output of each target model, select the business revenue with the largest value that reaches the target threshold;

[0126] 4. The candidate recommendation length corresponding to the target model that achieves the maximum business revenue value and reaches the target threshold is taken as the target recommendation length for that user.

[0127] For example, assume there are two target models: Target Model 1 (with a corresponding candidate recommendation length of 200) and Target Model 2 (with a corresponding candidate recommendation length of 300). If the business benefit output by Target Model 1 is greater than that output by Target Model 2, and the business benefit output by Target Model 1 reaches a target threshold, then the business benefit output by Target Model 1 is selected from Business Benefit 1 and Business Benefit 2 as the target business metric gain. Since the target business metric gain is output by Target Model 1, and the corresponding candidate recommendation length for Target Model 1 is 200, then 200 is used as the new recommendation length for this user, and this new recommendation length can be used as the target recommendation length.

[0128] In practical applications, if the business revenue output by each target model fails to reach the target threshold or the business revenue values ​​are the same, the original recommendation length for the user will be maintained, and product recommendations for the user will continue to be made using the original recommendation length.

[0129] 5. Recommend products to this user based on the target recommendation length.

[0130] In the aforementioned scheme, the business revenue with the largest value or that reaches the target threshold can be selected, and the candidate recommendation length corresponding to the target model that obtains such business revenue is taken as the target recommendation length for that user.

[0131] The above explanation uses two sets of target models as an example. In practical applications, the number of target models can also be other values, such as N sets of target models, where N is a positive integer greater than 2. Alternatively, N can also be 1.

[0132] In the above scheme, the target recommendation length is obtained based on the maximum business revenue and the target threshold. Obtaining the target recommendation length satisfies both the maximization of business effectiveness and the constraints of business computing power. In simpler terms, obtaining the target recommendation length is not only accurate but also maximizes business effectiveness, such as maximizing business revenue, while meeting the constraints of business computing power.

[0133] It is understandable that the fine-ranking stage in recommendation business application scenarios is a very computationally intensive stage. The length of data recommendation in the fine-ranking stage is directly proportional to the computational power consumption. By using the aforementioned technical solutions, the conversion efficiency between computational power and business results can be effectively improved.

[0134] The following describes the uplift models to be trained using the Multi-Treatment X-Learner algorithm.

[0135] Each group of uplift models to be trained consists of two regression models. Taking the first group of uplift models to be trained as an example, the first group of uplift models to be trained is used to process the business by changing the original recommended length (e.g., 100) to the candidate recommended length (e.g., 200).

[0136] The first set of uplift models to be trained consists of two regression models, μ0 and μ1.

[0137] μ0(x)=E[Y(0)|X=x] (1)

[0138] μ1(x)=E[Y(1)|X=x] (2)

[0139] Where x represents the traffic request data of the same user. Y(0) = 100 (original recommendation length); Y(2) = 200 (candidate recommendation length). E is a data operation representing mathematical expectation. μ0 corresponds to the original recommendation length and can be regarded as the regression model of the control group. μ1 corresponds to the candidate recommendation length and can be regarded as the regression model of the experimental group.

[0140] It is understandable that in each group of Uplift models to be trained, the historical traffic request data of the same user when the recommendation length is 200 is input into the control group regression model, and the result μ0(x) is obtained. 0 ) and the true label data Y when the candidate recommendation length is 200. 0 Perform residual calculations to obtain D. 0 The historical traffic request data of the same user when the recommendation length is 100 is input into the regression model of the experimental group. The true label data Y when the recommendation length is 100 is... 1 and the result obtained μ0(x 1 Perform residual calculations to obtain D. 1 .

[0141] D 1 =Y 1 -μ0(x 1 (3)

[0142] D 0 =μ0(x0 )-Y 0 (4)

[0143] The true label data for candidate recommendation length 200 and the true label data for candidate recommendation length 100 can be obtained by referring to the method for obtaining the label data in the training samples mentioned above.

[0144] The residual calculation result D 1 D 0 The labeled data, used as the basis for training the uplift model, is used to retrain the regression models τ1(x) and τ0(x). The regression models τ1(x) and τ0(x) are mathematically expressed as follows:

[0145] τ1(x)=E[D 1 |X=x] (5)

[0146] τ0(x)=E[D 0 |X=x] (6)

[0147] We take a weighted average of the regression models τ1(x) and τ0(x), where the weights g(x) can be functions of the input x, or g = 0.5, thus obtaining the first set of uplift models to be trained. Mathematically, this is expressed as τ(x):

[0148] τ(x)=g(x)*τ0(x)+(1-g(x))*τ1(x) (7)

[0149] The expressions for the uplift models to be trained in other groups are similar to those in formula (7). The aforementioned training of the uplift model to be trained is to train the τ1(x) and τ0(x) of that model. The training scheme is to determine the values ​​of τ1(x), τ0(x), and weight g(x) so that formula (7) can be applied to calculate the business indicator gain caused by the user processing the business from the original recommendation length (e.g., 100) to the candidate recommendation length (e.g., 200).

[0150] As can be seen from the aforementioned scheme, the training, target threshold determination, and target recommendation length acquisition processes disclosed herein are all white-boxed and highly interpretable. It can determine reasonable target business configuration parameters, such as the target recommendation length in the fine-ranking stage, based on traffic request data from different users, providing users with a personalized business experience and improving user experience. At the same time, it takes into account computing power consumption, maximizing business metric gains while meeting business computing power constraints, thus bringing substantial economic benefits.

[0151] In this disclosure, computing power consumption is simulated offline, which can predict computing power consumption offline. Based on the constraints of business computing power, the optimal uplift score threshold is selected from the candidate thresholds. The computing power constraints are well controllable, the process is transparent, and the implementation cost is low.

[0152] Using regression models as the base model, offline training avoids interference with online users during online training, reducing training costs while ensuring a better user experience.

[0153] It should be noted that the above algorithm is only one of them. In the field of causal inference, there are other algorithmic models that can be used as the gain model based on causal inference in this disclosure, including but not limited to: S-Learner, T-Learner, R-Learner, 2SLS, DR Learner, DRIV Learner, CEVAE, DragonNet, Uplift Tree, etc.

[0154] Because this disclosed embodiment can tailor target service configuration parameters for users, it can be regarded as a method for configuring personalized service parameters for users. Furthermore, it is a personalized recommendation method for service configuration parameters based on causal inference theory. Based on this solution, not only can personalized target service configuration parameters for different users be efficiently tailored, but also, under the effective constraint that overall computing power consumption is subject to business computing power constraints, the optimization of business indicators, such as maximizing business indicator gains, can be achieved.

[0155] The technical solution disclosed herein, based on target thresholds and business indicator gains, as well as target and computing power constraint thresholds, can automatically allocate optimal business configuration parameters to each search, recommendation, and other request on the Internet.

[0156] It is understood that the target model of this disclosure is a gain model based on causal inference. Benefiting from the causal inference method, the training process, the selection process of the target threshold, and the determination process of the target recommendation length of this disclosure can all be white-boxed. Compared with the black-box approach of using deep learning models in related technologies, it is more interpretable, transparent, and easy to port to various related fields.

[0157] This disclosure provides an apparatus for obtaining service configuration parameters, such as... Figure 7 As shown, the device includes:

[0158] The first acquisition unit 701 is used to acquire traffic request data;

[0159] The second acquisition unit 702 is used to input the feature data of traffic request data into at least one set of target models to obtain the output parameters of each set of target models; wherein, the output parameters of each set of target models represent the business indicator gain generated when the traffic request data is processed by changing from the original business configuration parameters corresponding to each set of target models to the candidate business configuration parameters corresponding to each set of target models.

[0160] The determining unit 703 is used to determine the target service configuration parameters from the original service configuration parameters and candidate service configuration parameters corresponding to each group of target models based on the service indicator gain of each group of target models. The target service configuration parameters are used to perform service processing on the object that generates the traffic request data.

[0161] In one optional embodiment, the determining unit 703 is configured to: determine a target business indicator gain from the business indicator gain outputs of each group of target models; wherein the target business indicator gain is at least one of the business indicator gain values ​​that are the largest among the business indicator gain values ​​output by each group of target models and the business indicator gain value that reaches the target threshold; and determine a target business configuration parameter from the original business configuration parameters and candidate business configuration parameters corresponding to each group of target models based on the target business indicator gain.

[0162] In one optional scheme, the target service configuration parameter is one of the following: the candidate service configuration parameter corresponding to the target model with the largest service indicator gain value among the service indicator gain values ​​output by each group of target models; the candidate service configuration parameter corresponding to the target model with the service indicator gain value that reaches the target threshold among the service indicator gain values ​​output by each group of target models; the candidate service configuration parameter corresponding to the target model with the largest service indicator gain value that reaches the target threshold among the service indicator gain values ​​output by each group of target models.

[0163] In an alternative embodiment, the determining unit 703 is configured to: in response to the fact that the business indicator gain output by each group of target models has not reached the target threshold, use the same original business configuration parameters as the target business configuration parameters.

[0164] In one alternative, the target threshold is obtained by filtering from at least one preset candidate threshold based on the historical business indicator gain obtained by the at least one set of target models for historical traffic request data, the computing power consumption under the historical business configuration parameters corresponding to the historical business indicator gain, and preset business computing power constraints.

[0165] In one optional scheme, the candidate service configuration parameter is the candidate recommendation length, and the original service configuration parameter is the original recommendation length; the service indicator gain is the service benefit generated by each group of target models when the traffic request data is searched or recommended using the original recommendation length corresponding to each group of target models instead of the candidate recommendation length corresponding to each group of target models.

[0166] In one alternative embodiment, the determining unit 703 is configured to: determine the target recommendation length from the original recommendation length and candidate recommendation length corresponding to each group of target models based on the business benefits of each group of target models.

[0167] In one alternative, the at least one set of target models is obtained by training each set of causal inference-based gain models with historical traffic request data and tag data obtained for each set of causal inference-based gain models.

[0168] The functions of each component in the service configuration parameter acquisition device of this disclosure embodiment can be found in the description of the service configuration parameter acquisition method, and will not be repeated here. Since the principle of solving the problem in the service configuration parameter acquisition device of this disclosure embodiment is similar to the aforementioned service configuration parameter acquisition method, the implementation process and implementation principle of the service configuration parameter acquisition device can be found in the aforementioned description of the implementation process and implementation principle of the related methods, and will not be repeated here.

[0169] According to embodiments of this disclosure, this disclosure also provides an electronic device, the electronic device including at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the aforementioned method for obtaining service configuration parameters.

[0170] For descriptions of processors and memory in electronic devices, please refer to [link to relevant documentation]. Figure 8 Description of the computing unit 801 and the storage unit 808.

[0171] According to embodiments of this disclosure, a non-transitory computer-readable storage medium storing computer instructions is also provided. These computer instructions are used to cause a computer to execute the aforementioned traffic control method and traffic control model training method. For a description of the computer-readable storage medium, please refer to... Figure 8 The relevant explanations are in the text.

[0172] According to embodiments of this disclosure, this disclosure also provides a computer program product, including a computer program that, when executed by a processor, achieves the acquisition of the aforementioned business configuration parameters. For a description of the computer program product, please refer to [link to relevant documentation]. Figure 8 The relevant explanations are in the text.

[0173] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0174] Figure 8 This is a block diagram of an electronic device used to implement an apparatus for obtaining service configuration parameters according to embodiments of the present disclosure. 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 processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0175] like Figure 8 As shown, the electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in ROM 802 or a computer program loaded into RAM 803 from storage unit 808. RAM 803 may also store various programs and data required for the operation of the electronic device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0176] Multiple components in electronic device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of displays, speakers, etc.; storage unit 808, such as disk, optical disk, or any device that can be used as a memory; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows electronic device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0177] The storage unit 808 in this embodiment may specifically be at least one of the following: portable computer disk, hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fiber, CD-ROM, optical storage device, magnetic storage device.

[0178] The computing unit 801 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, CPUs, graphics processing units (GPUs), artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processors, controllers, microcontrollers, etc. The computing unit 801 performs the various methods and processes described above, such as the method for obtaining service configuration parameters. For example, in some embodiments, the method for obtaining service configuration parameters may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the method for obtaining service configuration parameters described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform the method for obtaining service configuration parameters by any other suitable means (e.g., by means of firmware).

[0179] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, programmable 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.

[0180] 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 products that execute computer program code, such as processors or controllers of general-purpose computers, special-purpose computers, or other programmable data processing devices, 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, partially on a machine and partially on a remote machine as a standalone software package, or entirely on a remote machine or server. In the context of this disclosure, a machine-readable medium (storage medium) may 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 may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, 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, RAM, ROM, EPROM or flash memory, optical fibers, CD-ROMs, optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0181] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with a 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). The systems and techniques described herein can be implemented in computing systems that include back-end components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include front-end components (e.g., a user computer with a graphical user interface or web browser through which the user can interact with the implementations of the systems and techniques described herein), or computing systems that include any combination of such back-end components, middleware components, or front-end components. Components of the system can be interconnected via digital data communication (e.g., a communication network) of any form or medium. Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet. Computer systems can include clients and servers. Clients and servers are generally geographically distant from each other and typically interact via a communication network. Client-server relationships are created by computer programs running on respective computers and having client-server relationships with each other.

[0182] It should be understood that the various processes shown above can be used, with steps rearranged, added, or deleted. 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. 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 spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for obtaining a service configuration parameter, comprising: obtaining traffic request data; inputting feature data of the traffic request data into at least one set of target models to obtain output parameters of each set of target models; wherein the output parameters of each set of target models represent a service index gain amount generated by changing service processing of the traffic request data from using original service configuration parameters corresponding to each set of target models to using candidate service configuration parameters corresponding to each set of target models; determining a target service configuration parameter from the original service configuration parameters and the candidate service configuration parameters corresponding to each set of target models based on the service index gain amounts of each set of target models, the target service configuration parameter being used for service processing of an object generating the traffic request data.

2. The method of claim 1, wherein, The determining of the target service configuration parameter from the original service configuration parameters and the candidate service configuration parameters corresponding to each set of target models based on the service index gain amounts of each set of target models comprises: determining a target service index gain amount from the service index gain amounts output by each set of target models; wherein the target service index gain amount is a service index gain amount that is at least one of having a maximum value and reaching a target threshold among the service index gain amounts output by each set of target models; determining the target service configuration parameter from the original service configuration parameters and the candidate service configuration parameters corresponding to each set of target models based on the target service index gain amount.

3. The method of claim 1 or 2, wherein, The target service configuration parameter is one of: a candidate service configuration parameter corresponding to a target model that obtains a service index gain amount having a maximum value among the service index gain amounts output by each set of target models; a candidate service configuration parameter corresponding to a target model that obtains a service index gain amount reaching a target threshold among the service index gain amounts output by each set of target models; a candidate service configuration parameter corresponding to a target model that obtains a service index gain amount having a maximum value and reaching a target threshold among the service index gain amounts output by each set of target models.

4. The method of claim 3, wherein, The original service configuration parameters corresponding to each set of target models are the same; in response to none of the service index gain amounts output by each set of target models reaching the target threshold, using the same original service configuration parameter as the target service configuration parameter.

5. The method of claim 2, 3, or 4, wherein, The target threshold is obtained by screening at least one candidate threshold based on historical service index gain amounts obtained by the at least one set of target models for historical traffic request data, an amount of computing power consumption under historical service configuration parameters corresponding to the historical service index gain amounts, and a preset service computing power constraint condition.

6. The method of claim 1, wherein, The candidate service configuration parameter is a candidate recommendation length, and the original service configuration parameter is an original recommendation length; The service index gain amount is a service benefit generated by changing service search or service recommendation of the traffic request data from using original recommendation lengths corresponding to each set of target models to using candidate recommendation lengths corresponding to each set of target models.

7. The method of claim 6, wherein, The determining of the target service configuration parameter from the original service configuration parameters and the candidate service configuration parameters corresponding to each set of target models based on the service index gain amounts of each set of target models comprises: The target recommendation length is determined from the original recommendation length and the candidate recommendation length corresponding to each group of target models based on the business benefits of the groups of target models.

8. The method according to any one of claims 1 to 7, wherein, The at least one group of target models is obtained by training each group of causal inference-based gain models based on historical traffic request data and label data obtained for the groups of causal inference-based gain models.

9. An apparatus for obtaining a service configuration parameter, comprising: a first obtaining unit configured to obtain traffic request data; a second obtaining unit configured to input feature data of the traffic request data into at least one group of target models to obtain output parameters of the groups of target models, wherein the output parameters of the groups of target models represent business index gain amounts generated when the traffic request data is processed by using candidate service configuration parameters corresponding to the groups of target models instead of original service configuration parameters corresponding to the groups of target models; a determining unit configured to determine a target service configuration parameter from the original service configuration parameters and the candidate service configuration parameters corresponding to the groups of target models based on the business index gain amounts of the groups of target models, wherein the target service configuration parameter is used for processing the traffic request data.

10. The apparatus of claim 9, wherein, The determining unit is configured to: determine a target business index gain amount from the business index gain amounts output by the groups of target models, wherein the target business index gain amount is a business index gain amount that is at least one of having a maximum value and reaching a target threshold among the business index gain amounts output by the groups of target models; and determine the target service configuration parameter from the original service configuration parameters and the candidate service configuration parameters corresponding to the groups of target models based on the target business index gain amount.

11. The apparatus of claim 9 or 10, wherein, The target service configuration parameter is one of: a candidate service configuration parameter corresponding to a target model that has a maximum value among the business index gain amounts output by the groups of target models; a candidate service configuration parameter corresponding to a target model that has a business index gain amount reaching a target threshold among the business index gain amounts output by the groups of target models; and a candidate service configuration parameter corresponding to a target model that has a business index gain amount reaching a target threshold and having a maximum value among the business index gain amounts output by the groups of target models.

12. The apparatus of claim 11, wherein, The determining unit is configured to: in response to none of the business index gain amounts output by the groups of target models reaching the target threshold, use the same original service configuration parameter as the target service configuration parameter.

13. The apparatus of claim 10, 11 or 12, wherein, The target threshold is obtained by screening at least one candidate threshold based on a historical business index gain amount obtained by the at least one group of target models for historical traffic request data, an amount of computing power consumed under a historical service configuration parameter corresponding to the historical business index gain amount, and a preset service computing power constraint condition.

14. The apparatus of claim 9, wherein, The candidate service configuration parameter is a candidate recommendation length, and the original service configuration parameter is an original recommendation length. The service indicator gain amount is a service benefit generated by changing, for the traffic request data, service searching or service recommending using a candidate recommended length corresponding to each group target model to service searching or service recommending using an original recommended length corresponding to each group target model.

15. The apparatus of claim 14, wherein, The determining unit is configured to: determine, based on service benefits of each group target model, a target recommended length from the original recommended length and the candidate recommended length corresponding to each group target model.

16. The apparatus of any one of claims 9 to 15, wherein, The at least one group target model is obtained by training each group causal inference based gain model using historical traffic request data and label data obtained for each group causal inference based gain model. 17.An electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8. 18.A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method of any one of claims 1-8. 19.A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-8.

Images