Flow supporting method and device, electronic equipment and medium

Abstract

The embodiment of the invention discloses a flow supporting method and device, electronic equipment and a medium, and relates to the technical field of flow supporting, and a specific implementation mode of the method comprises the steps of determining an accumulated exposure threshold value associated with an article according to current state information of the article, and performing flow analysis on the article according to the accumulated exposure threshold value and a preset conversion threshold value, and carrying out potential identification on the article according to an analysis result so as to carry out flow support on the article according to a potential identification result. Dynamic flow analysis and accurate flow support in the object cold start stage are achieved, and the problems that the object flow distribution precision is low and the explosive incubation efficiency is poor are effectively solved.

Description

Technical Field

[0001] This application relates to the field of traffic support technology, specifically to a traffic support method, apparatus, electronic device, and medium. Background Technology

[0002] In e-commerce and content platforms, a large number of new products or content often face a "cold start" dilemma due to a lack of user behavior data, making it difficult to gain effective exposure, and high-quality hit products are easily buried. How to ensure that new products or content receive reasonable exposure and traffic allocation, achieve rapid traffic support, seamlessly transition to the stage of independent traffic competition, and at the same time incubate more high-quality hit products is a challenging problem.

[0003] Currently, traffic support solutions in related technologies mainly fall into two categories: one is model-based recommendation methods, which guide traffic allocation through item content similarity recall, tag preference matching, or training preference prediction models with limited data. The other is manual traffic allocation, which relies on expert experience or strategic planning to allocate fixed traffic to specific items through manual configuration, front-row placement, and other methods.

[0004] However, the relevant technologies have obvious drawbacks: model-based methods are limited by the sparsity of new item data, resulting in insufficient exposure after recall, low prediction accuracy, and limited traffic tilting and support effects; manual methods rely heavily on subjective experience, have poor stability, and are prone to traffic misallocation, such as overexposure of low-quality items and burying of high-quality items, thus causing overall traffic waste and revenue loss. Summary of the Invention

[0005] This application provides a flow support method, device, electronic device, and medium to solve the problems of low accuracy in item flow allocation and inaccurate identification of high-quality items in related technologies.

[0006] In a first aspect, embodiments of this application provide a method for traffic support, the method comprising: acquiring a set of items to be analyzed, including at least one item to be analyzed; determining a first cumulative exposure threshold associated with a first item to be analyzed based on the current status information of the first item to be analyzed, wherein the first item to be analyzed is any item to be analyzed in the set of items to be analyzed; performing traffic analysis on the first item to be analyzed based on the first cumulative exposure threshold and a preset conversion threshold, and obtaining analysis results; if the analysis results meet the potential identification conditions, determining the first item to be analyzed as a first item to be supported, and performing potential identification on the first item to be supported, and obtaining potential identification results; and if the potential identification results indicate high potential, providing traffic support to the first item to be supported.

[0007] In some embodiments, determining the first cumulative exposure threshold associated with the first item to be analyzed based on the current status information of the first item to be supported includes: determining the first decision exposure threshold of the first item to be supported based on the current status information of the first item to be analyzed; determining the traffic analysis period and the remaining time of traffic analysis for the first item to be analyzed; and determining the first cumulative exposure threshold based on the traffic analysis period, the remaining time of traffic analysis, and the first decision exposure threshold.

[0008] In some embodiments, traffic analysis is performed on a first item to be analyzed based on a first cumulative exposure threshold and a preset conversion threshold. The analysis results include: obtaining the cumulative exposure and cumulative conversion of the first item to be analyzed within the traffic analysis period; determining that the analysis result meets the potential identification conditions when the cumulative exposure is greater than or equal to the first cumulative exposure threshold and the cumulative conversion is greater than or equal to the preset conversion threshold, and / or when the cumulative conversion is greater than or equal to the preset conversion threshold; and determining that the analysis result does not meet the potential identification conditions when the cumulative exposure is greater than or equal to the first cumulative exposure threshold and the cumulative conversion is less than the preset conversion threshold, and / or when the cumulative conversion is less than the preset conversion threshold.

[0009] In some embodiments, potential identification of a first item to be supported, and obtaining potential identification results, includes: constructing a first potential outcome model based on the item characteristic information, support score label, and individual treatment effect value to be estimated of the first item to be supported; training a first preset estimation model based on the item characteristic information and support score label to obtain a propensity score model; training a second preset estimation model based on the item characteristic information and the first potential outcome model to obtain an outcome prediction model; estimating the individual treatment effect value to be estimated based on the item characteristic information, support score label, propensity score model, and outcome prediction model to obtain an individual treatment effect value; and identifying the potential of the first item to be supported based on the individual treatment effect value to obtain potential identification results.

[0010] In some embodiments, based on individual treatment effect values, the potential of the first item to be supported is identified, and the potential identification result includes: determining the effect estimation standard deviation corresponding to the individual treatment effect value; if the individual treatment effect value is greater than a preset treatment effect threshold and the effect estimation standard deviation is less than a preset effect estimation threshold, the potential identification result is determined to be high potential; if the individual treatment effect value is less than or equal to the preset treatment effect threshold, or the effect estimation standard deviation is greater than or equal to the preset effect estimation threshold, the potential identification result is determined to be low potential.

[0011] In some embodiments, when the potential identification result is high potential, providing traffic support to the first item to be supported includes: determining the first item to be supported as a first high-quality item when the potential identification result is high potential, and obtaining a set of high-quality items including the first high-quality item; determining the rate of change of the treatment effect of the first high-quality item based on the individual treatment effect value of the first high-quality item; obtaining a set of target high-quality items based on the rate of change of the treatment effect and a preset rate of change threshold; and ranking the at least one target high-quality item by score based on the individual treatment effect value of at least one target high-quality item in the set of target high-quality items, so as to provide traffic support to the at least one target high-quality item according to the score ranking result.

[0012] In some embodiments, the potential of a first item to be supported is identified to obtain a potential identification result. Then, the method further includes: determining a historical feedback dataset based on the analysis result and the potential identification result; updating the current status information of the first item to be analyzed based on the historical feedback dataset to obtain target status information; and determining a first updated cumulative exposure threshold associated with the first item to be analyzed based on the target status information, so as to perform traffic analysis on the first item to be analyzed according to the first updated cumulative exposure threshold.

[0013] Secondly, embodiments of this application provide a traffic support device, which includes: The acquisition unit is used to acquire a set of items to be analyzed, including at least one item to be analyzed. The determining unit is used to determine a first cumulative exposure threshold associated with the first item to be analyzed based on the current state information of the first item to be analyzed, wherein the first item to be analyzed is any item to be analyzed in the set of items to be analyzed; The traffic analysis unit is used to perform traffic analysis on the first item to be analyzed based on a first cumulative exposure threshold and a preset conversion threshold, and obtain the analysis results. The potential identification unit is used to determine the first item to be analyzed as the first item to be supported if the analysis results meet the potential identification conditions, and to identify the potential of the first item to be supported to obtain the potential identification result. The support unit is used to provide traffic support to the first item to be supported when the potential identification result is high potential.

[0014] Thirdly, embodiments of this application provide an electronic device, including: a processor and a memory for storing a computer program capable of running on the processor, wherein, when the processor runs the computer program, it performs the method described in any embodiment of the first aspect.

[0015] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the methods described in any embodiment of the first aspect.

[0016] This application provides a traffic support method, which involves: acquiring a set of items to be analyzed, including at least one item to be analyzed; determining a first cumulative exposure threshold associated with the first item to be analyzed based on its current state information, wherein the first item to be analyzed is any item in the set of items to be analyzed; performing traffic analysis on the first item to be analyzed based on the first cumulative exposure threshold and a preset conversion threshold to obtain analysis results; determining the first item to be analyzed as a first item to be supported if the analysis results meet the potential identification conditions, and performing potential identification on the first item to be supported to obtain potential identification results; and providing traffic support to the first item to be supported if the potential identification results indicate high potential. This method achieves dynamic traffic analysis and precise traffic support for items, avoids traffic waste or missed detection of potential items due to fixed traffic allocation standards by dynamically adapting exposure thresholds, and eliminates interference from items with falsely high conversion rates by combining precise potential identification. This effectively solves the problems of lack of targeted traffic allocation, low accuracy in identifying high-quality items, and poor efficiency in incubating blockbuster products in related technologies.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are merely embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort, and this application can be applied to other similar scenarios based on the provided drawings.

[0019] Figure 1 A flowchart illustrating a traffic support method provided in this application embodiment; Figure 2 A flowchart illustrating the second traffic support method provided in this application embodiment; Figure 3 A flowchart illustrating the training process for a decision threshold is provided in this application embodiment. Figure 4 A schematic diagram illustrating a process for dynamically adjusting the cumulative exposure threshold, provided in an embodiment of this application; Figure 5 A flowchart illustrating the third traffic support method provided in this application embodiment; Figure 6A schematic diagram of a potential identification process provided for an embodiment of this application; Figure 7 A flowchart illustrating the fourth traffic support method provided in this application embodiment; Figure 8 A schematic diagram illustrating a specific traffic support method provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of a flow support device provided in an embodiment of this application; Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. The described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that the terms "system," "device," "unit," and / or "module" used in this application are methods of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they can be replaced by other expressions.

[0022] Hereinafter, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first," "second," or "third" may explicitly or implicitly include one or more of that feature.

[0023] In e-commerce and content platforms, a large number of new products or content often face a "cold start" dilemma due to a lack of user behavior data, making it difficult to gain effective exposure, and high-quality hit products are easily buried. How to ensure that new products or content receive reasonable exposure and traffic allocation, achieve rapid traffic support (traffic support refers to providing specific items with additional, guaranteed exposure opportunities beyond regular recommendation and search traffic allocation mechanisms), seamlessly transition to a stage of self-competitive traffic, and simultaneously incubate more high-quality hit products, is a challenging problem.

[0024] Currently, there are two main types of traffic support solutions for the cold start phase of related technologies: one is model-based recommendation methods, such as item recall based on content similarity, which involves first identifying items liked by users and then recommending those items; or tag-based item recall, which involves first analyzing user behavior data, extracting user preference tags, and then selecting all items with these preference tags to push to users; or, based on limited performance data of items, trying different lightweight models and optimizing them to predict user preference scores for items. The other type is manual traffic allocation, which relies on expert experience or strategic planning to allocate fixed traffic to specific items through manual configuration, front-row push, etc.

[0025] However, the relevant technologies have obvious drawbacks: model-based methods are limited by the sparsity of new item data, resulting in insufficient exposure after recall, low prediction accuracy, and limited traffic tilting and support effects; manual methods rely heavily on subjective experience, have poor stability, and are prone to traffic misallocation, such as overexposure of low-quality items and burying of high-quality items, thus causing overall traffic waste and revenue loss.

[0026] To address the problems in related technologies, this application proposes a traffic support method. This method determines a first cumulative exposure threshold based on the current status information of a first item to be analyzed. Then, it performs traffic analysis on the first item based on the first cumulative exposure threshold and a preset conversion threshold. If the analysis results meet the potential identification criteria, the first item is identified as a first item to be supported. The first item to be supported is then subjected to potential identification. If the potential identification result indicates high potential, traffic support is provided to the first item to be supported. This achieves dynamic and adaptive traffic allocation for items. The dynamically adapted exposure threshold avoids traffic waste or missed potential items caused by fixed traffic allocation standards. Combined with accurate potential identification, it eliminates interference from false high conversion rates, effectively solving the problems of lack of targeted traffic allocation, inaccurate prediction of item preferences, low accuracy in identifying high-quality items, and poor efficiency in incubating blockbuster products in related technologies.

[0027] The following section provides a detailed description of a traffic support method provided in this application, with reference to the accompanying drawings.

[0028] Figure 1 A flowchart illustrating a traffic support method provided in an embodiment of this application is shown. Figure 1 As shown, the traffic support method includes steps 101-105.

[0029] Step 101: Obtain a set of items to be analyzed, including at least one item to be analyzed.

[0030] In the embodiments of this application, the items to be analyzed refer to newly listed items in the cold start phase, lacking sufficient user behavior data, and requiring traffic testing and evaluation, such as new products on e-commerce platforms, new videos or articles on content platforms, etc. The set of items to be analyzed is the total set of new items defined by the platform that are waiting for traffic analysis, which may include newly listed items on the same day, manually submitted items for retesting (i.e., items whose historical analysis results do not meet the potential identification requirements), and items for delayed analysis (i.e., items that are arranged to enter the traffic analysis process later due to specific strategy considerations), etc.

[0031] The cold start phase refers to the initial stage after new items (such as goods, content, and services) are listed on the platform. Due to the lack of sufficient user behavior data (such as clicks, favorites, purchases, and comments), it is impossible to accurately match user needs and obtain stable organic traffic through traditional recommendation models.

[0032] It should be noted that for manually reported items to be retested, their historical flow analysis data must be cleared to eliminate the interference of historical analysis data on the results of this analysis.

[0033] Specifically, the system typically scans the item database via scheduled tasks (such as every morning), adding items that meet the definition of "items to be analyzed" to a specific storage (such as a Redis collection or database table), forming the set of items to be analyzed for the day. This process can be integrated with a manual reporting mechanism, allowing operations personnel to directly add specific strategic new products to this set.

[0034] Step 102: Based on the current status information of the first item to be analyzed, determine the first cumulative exposure threshold associated with the first item to be analyzed. The first item to be analyzed is any item in the set of items to be analyzed.

[0035] In the embodiments of this application, the current state information is used to describe a set of dynamic characteristics of the item's own attributes and its environment, including at least: category characteristics (such as the category to which it belongs, the average conversion cycle of items in that category, etc.); historical performance (the best threshold range of similar items in the past, the false judgment rate, etc.); and traffic distribution characteristics (such as the current peak traffic period of the platform, user activity, etc.). The first cumulative exposure threshold is a target value of total exposure that needs to be achieved in the entire traffic analysis period for the first item to be analyzed, and this threshold will be dynamically adjusted as the traffic analysis progresses.

[0036] Specifically, the current state information is taken as input and passed to a pre-trained Deep Reinforcement Learning (DRL) agent. The DRL agent outputs a basic exposure threshold suggestion value based on its internalized strategy. Then, the basic exposure threshold is adjusted in combination with the time urgency factor to generate the final personalized cumulative exposure threshold.

[0037] The time urgency factor is calculated based on the flow analysis cycle of the first item to be analyzed and the remaining time of the flow analysis activity. It is a dynamically adjusted scaling factor used to adaptively regulate the exposure threshold within the flow analysis cycle in order to balance testing efficiency and resource allocation fairness.

[0038] Step 103: Based on the first cumulative exposure threshold and the preset conversion threshold, perform traffic analysis on the first item to be analyzed to obtain the analysis results.

[0039] In the embodiments of this application, the preset conversion threshold refers to a pre-set minimum conversion index for an item, used as the minimum performance standard to judge the quality of an item, such as a minimum click-through rate or conversion rate. It is typically set to a small value to eliminate poorly performing items in advance.

[0040] The analysis results are based on the performance of the items at the end of the flow analysis cycle or when the conditions are triggered early, and the performance of the items to be analyzed is classified into two categories: those that meet the potential identification conditions and those that do not meet the potential identification conditions.

[0041] Specifically, based on the first cumulative exposure threshold and the preset conversion threshold, traffic analysis is performed on the first item to be analyzed to determine whether the cumulative exposure and cumulative conversion of the first item to be analyzed within the entire traffic analysis period have reached the threshold. If the threshold is reached, the first item to be analyzed is considered to meet the potential identification conditions; otherwise, the first item to be analyzed is considered not to meet the potential identification conditions.

[0042] If the analysis results indicate that the potential identification criteria are not met, the first item to be analyzed will be removed from this traffic support process and will not proceed to subsequent potential identification and traffic support stages. Alternatively, the reasons for not meeting the potential identification criteria will be analyzed, and the item will be retested in the next round by adjusting the price, optimizing the cover material, etc., to accumulate more experience. If the analysis results of the first item to be analyzed indicate that the potential identification criteria are met, then proceed to step 104 for potential identification.

[0043] Step 104: If the analysis results meet the potential identification conditions, determine the first item to be analyzed as the first item to be supported, and perform potential identification on the first item to be supported to obtain the potential identification results.

[0044] In the embodiments of this application, meeting the potential identification criteria means that the cumulative exposure and cumulative conversion of the first item to be analyzed reach a threshold throughout the entire traffic analysis period. Each item that meets the potential identification criteria is further screened as an item to be supported in order to identify items with high potential, which are the target objects for subsequent precise traffic support.

[0045] This application uses a causal inference method to eliminate the influence of other confounding factors (such as the quality of the product itself and promotional activities during the same period) and assess the net improvement effect of the action of "providing support traffic" on the product's performance indicators, thereby identifying truly high-potential products.

[0046] Specifically, this application estimates the individual treatment effect (the difference in conversion rate between items receiving traffic support and those not receiving it) of the items to be supported using a dual machine learning approach. In one example, a gradient boosting tree model is first used to fit the exposure tendency (i.e., the tendency to receive traffic support) of the items to be supported, and a neural network model is used to fit the natural conversion rate, natural order volume, etc., of the items to be supported without additional traffic support. Then, the individual treatment effect of the items to be supported is calculated based on the output of these two models. Finally, the estimated individual treatment effect value is used to identify the potential of the items to be supported, thereby identifying whether the items to be supported have high potential.

[0047] Step 105: If the potential identification result is high potential, provide traffic support to the first item to be supported.

[0048] In the embodiments of this application, traffic support refers to providing additional, guaranteed exposure opportunities for specific items beyond conventional traffic allocation mechanisms such as recommendations and searches.

[0049] Specifically, firstly, items deemed high-potential for support are added to a separate pool of high-quality items (such as a Redis sorted set). These items are then sorted from highest to lowest based on their individual processing effect value, and their subsequent performance is continuously monitored. Next, a traffic-priority strategy is implemented for the items in the high-quality item pool, including: placing items from the high-quality new product pool at the top of the recommendation lists on various applications to increase their exposure priority; providing fixed display in specific locations within the application (such as "New Products Zone" or "Editor's Picks"); and adding exclusive tags such as "New" or "Potential Hit" to items during display to enhance user attention.

[0050] In summary, the traffic support method proposed in this application involves: obtaining a set of items to be analyzed, including at least one item to be analyzed; determining a first cumulative exposure threshold associated with the first item to be analyzed based on its current state information; performing traffic analysis on the first item to be analyzed based on the first cumulative exposure threshold and a preset conversion threshold to obtain analysis results; determining the first item to be analyzed as the first item to be supported if the analysis results meet the potential identification conditions, and performing potential identification on the first item to be supported to obtain potential identification results; and providing traffic support to the first item to be supported if the potential identification results indicate high potential. This method achieves dynamic traffic analysis and precise traffic support during the cold start phase of items. By dynamically adapting the exposure threshold, it avoids traffic waste or missed detection of potential items caused by fixed traffic allocation standards. Combined with precise potential identification, it eliminates interference from false high conversion rates, effectively solving the problems of lack of targeted traffic allocation, low accuracy in identifying high-quality items, and poor efficiency in incubating blockbuster products in related technologies.

[0051] As one possible implementation method, Figure 2 A flowchart of the second traffic support method is shown. Based on the above embodiment, a first cumulative exposure threshold associated with the first item to be analyzed is determined based on the current status information of the first item to be supported, including the following steps: Step 201: Based on the current status information of the first item to be analyzed, determine the first decision exposure threshold for the first item to be supported.

[0052] In the embodiments of this application, the first decision exposure threshold is a baseline value of exposure that the current item to be analyzed should achieve under ideal or standard conditions, directly output by the DRL agent based on the current state information of the first item to be analyzed. The DRL agent is pre-trained based on historical interaction data from the cold start traffic analysis scenario of the item using the Deep Deterministic Policy Gradient (DDPG) algorithm. The specific training process is as follows: Figure 3 As shown, Figure 3 This is a flowchart illustrating the training process for a decision threshold, as provided in an embodiment of this application.

[0053] Reference Figure 3The DRL agent first interacts with the item traffic analysis environment (Env) to obtain the current state of the item to be analyzed (such as category characteristics, historical analysis performance, traffic distribution characteristics, etc.) and outputs an initial decision action (i.e., candidate decision exposure threshold). The environment generates a transition sample containing "current state s, action a, reward r, and new state s'" based on the action feedback reward (e.g., "positive reward for identifying high-potential items, negative reward for excessive traffic waste") and the new state, and stores it in the experience replay pool. Samples are randomly sampled from the experience replay pool and input into the Critic network for training. The evaluation accuracy of "action value (Q value, i.e., the value used to evaluate the exposure threshold)" is optimized by minimizing the TD (temporal difference) error. At the same time, the samples are input into the Actor network for training. By changing the policy gradient direction, the action (exposure threshold) output by the network can maximize the Q value evaluated by the Critic network. The closed-loop process of environment interaction, sample storage, and network training is repeated until the decision exposure threshold output by the DRL agent stably reaches the preset training target in the historical traffic analysis scenario, thus completing the training of the DRL agent.

[0054] Specifically, the system will collect the current status information of the first item to be analyzed. (Including category characteristics, historical performance analysis, traffic distribution characteristics, etc.) Input the pre-trained DRL agent, and the policy network of the DRL agent outputs the basic exposure decision value for the item, i.e., the first decision exposure threshold, based on the input state information. For example, for a high-end product, the model may output a higher exposure threshold (such as 10000); for a regular new product, it may output a lower exposure threshold (such as 3000).

[0055] Step 202: Determine the flow analysis cycle and remaining flow analysis time for the first item to be analyzed.

[0056] In the embodiments of this application, the traffic analysis period refers to the complete traffic analysis duration set for the first item to be analyzed. The analysis period for different categories of items can be configured differently (e.g., 3 days for new product testing for clothing, and 7 days for new product testing for electronic products). The remaining traffic analysis time refers to the remaining time from the current time point until the end of the traffic analysis period for the item, which is a dynamically changing amount of time.

[0057] Step 203: Determine the first cumulative exposure threshold based on the traffic analysis cycle, the remaining time of traffic analysis, and the first decision exposure threshold.

[0058] In this embodiment, the first cumulative exposure threshold refers to the maximum exposure limit of the item within the entire analysis period, determined after integrating item attribute features and time factors. It is the core quantitative indicator for determining whether the item has completed traffic analysis. The formula for calculating the cumulative exposure threshold is as follows:

[0059] in, Let DRL be the decision function of the agent; The decision-making exposure threshold; This is the current state space; This represents the time urgency factor. Remaining time for traffic analysis; For flow analysis cycle; The time urgency factor is a dynamically adjusted scaling factor.

[0060] It is understandable that dividing the first cumulative exposure threshold by the number of days in the traffic analysis period will yield the first day's first-level exposure threshold for the first item to be analyzed.

[0061] Optionally, this application calculates real-time data such as the cumulative exposure and conversion of the first item to be analyzed on a daily basis through an online real-time stream, and reads the previously accumulated daily exposure and conversion data of the first item to be analyzed from the daily set of items to be analyzed. The real-time data and daily data are then integrated and input into the DRL agent, enabling dynamic optimization of the cumulative exposure threshold. Figure 4 As shown, Figure 4 This is a schematic diagram illustrating a process for dynamically adjusting the cumulative exposure threshold, as provided in an embodiment of this application. The specific process is as follows: Reference Figure 4 First, the real-time and daily data of the items to be analyzed are cleaned and standardized, and integrated to form state data that can be recognized by the DRL agent. Then, the integrated state data is input into the DRL agent to complete the encoding and extraction of state features. The policy network of the DRL agent adjusts the cumulative exposure threshold in real time based on the current state data. According to the adjusted cumulative exposure threshold, the real-time exposure of the items to be analyzed is dynamically controlled. Combining the actual conversion effect and traffic utilization efficiency after exposure control, a reward signal is generated and fed back to the DRL agent for subsequent strategy optimization.

[0062] Through the above process, the DRL agent updates the cumulative exposure threshold periodically (e.g., every 5 minutes), thereby achieving dynamic adjustment of the cumulative exposure threshold.

[0063] After determining the first cumulative exposure threshold, this application needs to perform traffic analysis on the first item to be analyzed based on the first cumulative exposure threshold and the preset conversion threshold to obtain the analysis results.

[0064] The preset conversion threshold can be set manually based on historical experience, and its specific value can be configured differently based on factors such as the category of the item and the platform's operational needs. It is not limited in the embodiments of this application.

[0065] The analysis results of this application include two categories: those that meet the potential identification conditions and those that do not. The specific logic for determining whether the first item to be analyzed meets the potential identification conditions based on the first cumulative exposure threshold and the preset conversion threshold is as follows: In the embodiments of this application, the cumulative exposure and cumulative conversion of the first item to be analyzed during the traffic analysis period are first obtained through online real-time streaming.

[0066] Among them, cumulative exposure is the total number of exposures obtained by the first item to be analyzed within the traffic analysis period; cumulative conversion is the effective conversion data generated by the first item to be analyzed within the traffic analysis period, such as the number of completed orders and click-through rate.

[0067] In the embodiments of this application, when the cumulative exposure is greater than or equal to a first cumulative exposure threshold and the cumulative conversion is greater than or equal to a preset conversion threshold, and / or the cumulative conversion is greater than or equal to a preset conversion threshold, the analysis result is determined to meet the potential identification conditions.

[0068] In the embodiments of this application, if the cumulative exposure is greater than or equal to a first cumulative exposure threshold and the cumulative conversion is less than a preset conversion threshold, and / or the cumulative conversion is less than a preset conversion threshold, the analysis result is determined to be that the potential identification condition is not met.

[0069] Optionally, during the entire traffic analysis period, if the cumulative exposure of the first item to be analyzed is less than the first cumulative exposure threshold or the first item to be analyzed is taken down and becomes invalid, the first item to be analyzed is also considered not to meet the potential identification conditions.

[0070] Optionally, this application implements an adaptive weighting strategy for the first item to be analyzed whose cumulative exposure is approaching a first cumulative exposure threshold, in order to optimize traffic allocation and improve the overall efficiency of the traffic analysis process. The specific adaptive weighting coefficients are as follows:

[0071] in, , This is a gain parameter that can be adjusted according to the characteristics of the category. This represents the current cumulative exposure of the first item to be analyzed.

[0072] In summary, this application generates a basic decision-making exposure threshold based on the real-time status information of the item to be analyzed, and then obtains a dynamic cumulative exposure threshold for controlling the entire traffic analysis process based on the traffic analysis cycle and the remaining time of traffic analysis. This effectively solves the problems of traffic waste and missed detection of potential items caused by traditional fixed threshold or simple rule methods, enabling the traffic allocation strategy to adapt to both item characteristics and time pressure, thereby maximizing traffic analysis efficiency and the success rate of discovering top-selling items with limited resources.

[0073] As one possible implementation method, Figure 5 A flowchart of the third traffic support method is shown. Based on the above embodiments, the potential of the first item to be supported is identified, and the potential identification result is obtained, including the following steps: Step 301: Based on the item characteristic information, support score label and individual treatment effect value to be estimated of the first item to be supported, construct the first potential outcome model.

[0074] In the embodiments of this application, for each item that meets the potential identification criteria, dual machine learning is used to estimate its individual treatment effect to determine whether it is a high-quality item, so as to provide subsequent traffic support.

[0075] Among them, the item characteristic information refers to the core attribute data of the first item to be supported, which is a multi-dimensional feature vector, typically including: Static characteristics of products: such as category, price, brand, listing time, and material quality score; Contextual features: such as the time of listing, whether it is peak season, and the number of competing items at the same time; Traffic analysis phase performance: Initial data accumulated during the traffic analysis phase, such as the initial trends of impressions, click-through rate, and conversions.

[0076] The support score label is a binary label indicating whether an item received traffic support, with a value of "1" (indicating traffic support received) or "0" (indicating no traffic support received). It is used to distinguish the item's performance under different intervention states. The individual treatment effect value to be estimated is an unknown quantity to be solved, representing the incremental causal effect of support traffic on the item's performance.

[0077] This application employs a causal inference method to eliminate the influence of other confounding factors (such as the quality of the product itself and concurrent promotional activities) and assess the net improvement effect of the action of "providing support traffic" on product performance indicators, thereby identifying truly high-potential products. Causal inference can identify "causality" (i.e., whether receiving support traffic can truly improve the long-term conversion of the product). Its core is to construct a counterfactual reasoning framework, that is, to construct a potential outcome model, as follows:

[0078] in, T To support scoring tags ( T =1 / 0); Items awaiting support i In state T Potential observational results (such as click-through rate, conversion rate); Items awaiting support i Item feature vector; These are the characteristic coefficients of the items; The individual treatment effect value to be estimated; This is due to random error (uncontrollable factors, such as random clicks by the user).

[0079] What this application needs to estimate is... This is to assess how much improvement the support traffic can bring to the item after stripping away its inherent characteristics.

[0080] Step 302: Based on the item feature information and support score labels, train the first preset estimation model to obtain the propensity score model.

[0081] In the embodiments of this application, a dual machine learning method is used to estimate... The first preset estimation model refers to a machine learning model used to predict the probability of an item receiving traffic support. This application takes the gradient boosting tree model (which is good at handling high-dimensional discrete features and nonlinear relationships) as an example.

[0082] In one example, this application uses item feature vectors as features and support score labels as labels to construct a training dataset. , T Using this training dataset as input to the gradient boosting tree model for training, the following propensity score model is obtained:

[0083] The trained propensity score model can predict the likelihood of an item receiving traffic support based on its characteristic information.

[0084] Step 303: Based on the item feature information and the first potential outcome model, train the second preset estimation model to obtain the outcome prediction model.

[0085] In the embodiments of this application, the second preset estimation model refers to a machine learning model used to predict the natural conversion result of an item when it does not receive traffic support. This solution uses a neural network model (which is good at capturing complex interaction relationships between features and is adapted to the need for accurate prediction of natural conversion results).

[0086] In one example, this application constructs a training dataset using item feature vectors as features and potential outcome models as labels. , Using this training dataset as input to the neural network model for training, the resulting prediction model is as follows:

[0087] The trained predictive model can predict the natural performance of an item when it does not receive traffic support, based on the item's characteristic information.

[0088] Step 304: Based on the item feature information, support score labels, propensity score model and outcome prediction model, estimate the individual treatment effect value to be estimated, and obtain the individual treatment effect value.

[0089] In the embodiments of this application, the application optimizes the estimation of the individual treatment effect value by stripping away the bias in support allocation using propensity score residuals and by stripping away the influence of inherent characteristics of items using result prediction residuals, thus obtaining an unbiased individual treatment effect value. The formula for calculating the propensity score residual is as follows: residual_T = T - g_model1.predict( X i ) The formula for calculating the predicted residual is as follows: residual_Y = Y i (1) - m_model1.predict( X i ) The individual treatment effect value can be obtained by dividing the outcome prediction residual by the propensity score residual, as follows:

[0090] in, Individual treatment effect value.

[0091] Step 305: Based on the individual treatment effect value, identify the potential of the first item to be supported and obtain the potential identification result.

[0092] In the embodiments of this application, this application determines whether the items to be supported are of high potential and worthy of long-term key support based on the magnitude and reliability of the estimated individual treatment effect value.

[0093] This application identifies the potential of a first category of items to be supported based on individual treatment effect values. The potential identification results include: determining the standard deviation of the effect estimate corresponding to the individual treatment effect value; determining the potential identification result as high potential when the individual treatment effect value is greater than a preset treatment effect threshold and the standard deviation of the effect estimate is less than a preset effect estimate threshold; and determining the potential identification result as low potential when the individual treatment effect value is less than or equal to the preset treatment effect threshold, or the standard deviation of the effect estimate is greater than or equal to the preset effect estimate threshold.

[0094] In the embodiments of this application, the preset treatment effect threshold is typically set to the higher quantile (e.g., the 75th quantile) of the individual treatment effect values ​​of all items to be supported, ensuring that only the top items with the strongest effects are selected. The preset effect estimation threshold is used to ensure that the estimation results are reliable enough and to avoid misjudging highly fluctuating noise as high potential. The preset effect estimation threshold can be set manually according to the actual situation, and is not limited in the embodiments of this application. This application uses 0.2 as an example.

[0095] In one example, the criteria for determining high-potential items in this application are as follows:

[0096] in, The 75th percentile of the individual treatment effect value of all items in need of support; Estimate the standard deviation of the effect (less than 0.2 indicates high confidence); Indicates items requiring support i It has high potential. Indicates items requiring support i It has low potential.

[0097] To further explain the entire process of potential identification, refer to... Figure 6 , Figure 6 This is a schematic diagram of a potential identification process provided in an embodiment of this application.

[0098] Reference Figure 6 Products that have successfully entered the support phase (corresponding to the items that meet the potential identification criteria in this application) and manually submitted items (referring to items actively submitted by operators and whose potential needs to be assessed) require further causal effect evaluation (i.e., estimating the individual treatment effect value of the items). Based on the item's characteristic information (static characteristics, contextual characteristics, performance during the traffic analysis period, etc.) and support score tags (whether traffic intervention was received), the individual treatment effect value of the items to be supported is accurately estimated by constructing a potential outcome model, training a propensity score model, and an outcome prediction model. Based on the individual treatment effect value and the standard deviation of the effect estimate of the items to be supported, and compared with preset thresholds (i.e., preset treatment effect thresholds and preset effect estimation thresholds), the potential value of the items to be supported is calculated. Items with high potential are added to the high-quality item pool as the core targets for subsequent traffic support and resource allocation.

[0099] In some embodiments, after identifying the potential of the first item to be supported and obtaining the potential identification result, the method further includes: determining a historical feedback dataset based on the analysis result and the potential identification result; updating the current status information of the first item to be analyzed based on the historical feedback dataset to obtain target status information; and determining a first updated cumulative exposure threshold associated with the first item to be analyzed based on the target status information, so as to perform traffic analysis on the first item to be analyzed according to the first updated cumulative exposure threshold.

[0100] In the embodiments of this application, a feedback dataset is constructed based on the analysis and identification results of the traffic analysis stage and the support stage, and the DRL agent is retrained periodically to improve the accuracy of the cumulative exposure threshold.

[0101] The analysis results include two categories: "meets the potential identification criteria" and "does not meet the potential identification criteria" (including special cases such as insufficient cumulative exposure or product removal and invalidation); the potential identification results include two types: "high potential" and "low potential".

[0102] The historical feedback dataset is a structured dataset that records the comparison between the judgment results and the final truth of items throughout the entire process from "traffic analysis" to "potential identification" over a period of time. It includes positive feedback samples (items that meet the potential identification conditions and are judged as high potential), false positive feedback samples (items that meet the potential identification conditions but are judged as low potential, which is a misjudgment), and false negative feedback samples (items that do not meet the potential identification conditions but are subsequently judged as high potential after manual retesting).

[0103] Specifically, this application first updates the current state information of the first item to be analyzed based on the historical feedback dataset, that is, in the current state space. By incorporating features such as "historical positive feedback rate" and "historical misjudgment rate" into the data (including category characteristics, historical analysis performance, traffic distribution characteristics, etc.), the target state space is obtained. (Composed of target state information). Then, update the original reward function used to train the DRL agent (original reward function = best-selling product recognition accuracy × traffic utilization rate) to obtain the target reward function (used to guide the DRL agent to continuously optimize the best-selling product recognition accuracy), as follows:

[0104] in, The original reward function; This represents the number of positive feedback samples in the current training batch. This represents the number of false positive feedback samples in the current training batch. This represents the number of false negative feedback samples in the current training batch. , and These are weighting coefficients, which can be adjusted through cross-validation.

[0105] Finally, every so often (e.g., weekly), the DRL agent is trained again using the target state space and target reward function. This allows the DRL agent to learn to increase the exposure threshold for high-potential items, which can reduce false negatives. For items that are prone to misjudgment, such as items with large conversion fluctuations, a more stringent conversion rate check is introduced to reduce false negatives.

[0106] In summary, this application employs a dual machine learning approach to perform causal inference on items seeking support, eliminating the influence of confounding factors such as the product's inherent characteristics and the external environment. This allows for the accurate estimation of the true individual treatment effect of traffic support on each item, and based on this individual treatment effect, high-potential items are rigorously selected for focused support. Simultaneously, the traffic analysis results and potential identification results are integrated into historical feedback data, iteratively updating item status information and optimizing cumulative exposure thresholds to form a closed loop. This ensures the scientific rigor of potential identification and enables dynamic and precise allocation of traffic resources, further improving the success rate of creating blockbuster products and overall traffic utilization efficiency.

[0107] As one possible implementation method, Figure 7 A flowchart of the fourth traffic support method is shown. Based on the above embodiments, when the potential identification result is high potential, traffic support is provided to the first item to be supported, including the following steps: Step 401: If the potential identification result is high potential, determine the first item to be supported as the first high-quality item, and obtain a set of high-quality items including the first high-quality item.

[0108] In the embodiments of this application, whenever the potential identification of a batch of items is completed, the system adds all items marked as "high potential" and their key information (such as product identifier, category, estimated individual treatment effect value, decision time, etc.) to a specific storage structure to form a set of high-quality items.

[0109] Step 402: Based on the individual treatment effect value of the first premium item, determine the rate of change of the treatment effect of the first premium item.

[0110] In the embodiments of this application, the treatment effect change rate is used as an indicator to measure the potential sustainability of a high-quality item. It is calculated by comparing the relative change of the current individual treatment effect value of the high-quality item with its historical effect value at a certain historical moment. This reflects whether the benefits gained from supporting the high-quality item are increasing, stable, or declining. High-quality items with declining performance are promptly removed from the high-quality item set to ensure that traffic is tilted towards items with "stable gains and sustainable potential." High-quality items i The formula for calculating the rate of change of the treatment effect is as follows:

[0111] in, For high-quality items i The individual treatment effect value at the current calculation time; For high-quality items i exist Individual treatment effect values ​​from 1 day ago.

[0112] Step 403: Based on the rate of change of the treatment effect and the preset rate of change threshold, obtain the target set of high-quality items.

[0113] In the embodiments of this application, the preset rate of change threshold is a negative critical value used to judge the severity of the attenuation of the individual treatment effect. It can be configured according to actual needs and is not limited in the embodiments of this application; -0.3 is used as an example. This application performs elimination operations on high-quality items based on the rate of change of the treatment effect and the preset rate of change threshold to ensure the accuracy and overall efficiency of subsequent traffic support. Specifically, the elimination rules of this application are as follows: When high-quality items i The rate of change of the treatment effect is less than the preset rate of change threshold (e.g. When, high-quality items i Remove from the set of high-quality items, and the remaining high-quality items that do not meet the elimination rules constitute the current target set of high-quality items.

[0114] Step 404: Based on the individual treatment effect value of at least one target high-quality item in the target high-quality item set, sort the at least one target high-quality item by score, and provide traffic support to the at least one target high-quality item according to the score sorting result.

[0115] In the embodiments of this application, all target high-quality items in the target high-quality item set are sorted in descending order according to a comprehensive score. This score is usually used directly or derived from the individual treatment effect value (such as the individual treatment effect value itself, the weighted combination of the individual treatment effect value and recent sales, etc.), representing the priority of traffic support for them at the current moment.

[0116] In one example, the target high-quality items are inserted into the original sorting results of each supported application according to their scores, and traffic support is provided for the target high-quality new products. At the same time, when the target high-quality items are displayed, visual tags such as "potential best-seller", "best-selling new product" and "New" can be added to strengthen the traffic support for the target high-quality items.

[0117] In summary, this application, after obtaining a collection of high-quality items, implements continuous causal effect monitoring and a dynamic survival-of-the-fittest mechanism for high-potential items, achieving precise iteration and efficient utilization of traffic support resources, and significantly improving the overall return on supported traffic and the success rate of incubating blockbuster products.

[0118] In one example, to help better understand the traffic support scheme provided in this application, such as Figure 8 As shown, Figure 8 This diagram illustrates a specific traffic support method provided in an embodiment of this application. The traffic support method of this application is divided into a traffic analysis stage and a potential identification stage.

[0119] Reference Figure 8 During the traffic analysis phase, the system acquires a daily set of new products to be tested (corresponding to the set of items to be analyzed in this application). These items receive exposure in the new product testing (i.e., traffic analysis) block. The online real-time traffic engine monitors and accumulates data such as exposure, clicks, and conversions for each item in real time through interactive calculations. Simultaneously, the DRL agent dynamically adjusts the exposure threshold strategy for each item based on its real-time status and makes dynamic removal decisions based on preset rules (such as reaching a threshold or having a low conversion rate). For example, items with daily cumulative exposure exceeding the daily exposure threshold, cumulative exposure exceeding the cumulative exposure threshold, or cumulative conversion below the conversion threshold are removed from the new product testing set. Furthermore, in each new product testing block, multiple auxiliary recommendation models can be integrated to improve accuracy. For instance, user acceptance models and user item preference models can be overlaid. Based on the established user interest in new products, specific new products that best match the user's historical interests are further selected for exposure, thereby improving the overall efficiency of the new product testing traffic. At the end of each daily traffic analysis cycle, the system determines the status of all items to be analyzed, including: incomplete, failed, and successful (where "success" corresponds to "meets potential identification conditions" in this application, and "failure" and "incomplete" correspond to "does not meet potential identification conditions" in this application). Finally, the system comprehensively considers newly added products of the day, as well as items that need to be retested or have their testing delayed, to generate a set of items to be analyzed the next day, thus initiating a new round of traffic analysis.

[0120] During the potential identification phase, for items with successful traffic analysis results, the system conducts in-depth analysis based on causal inference methods to identify truly high-potential, high-quality new products. These identified high-quality new products will be pushed to various business applications such as search, recommendation, and advertising, receiving focused traffic support through methods such as weighted ranking, tagging, and dedicated display areas.

[0121] This application also includes a collaborative optimization phase, in which the analysis and identification results of historical high-quality new products, as well as data such as the rate of change of causal effects during the potential identification period, are used as key feedback signals and input into the training process of the DRL agent for updating and optimizing the strategy model. This enables the threshold decision strategy in the traffic analysis phase to be continuously optimized, and to more accurately predict and screen potential new products.

[0122] To achieve the above embodiments, this application also provides a flow support device. Figure 9 This is a schematic diagram of the structure of a flow support device 900 provided in an embodiment of this application. Figure 9 As shown, the device includes: Acquisition unit 910 is used to acquire a set of items to be analyzed, including at least one item to be analyzed; The determining unit 920 is used to determine a first cumulative exposure threshold associated with the first item to be analyzed based on the current state information of the first item to be analyzed, wherein the first item to be analyzed is any item to be analyzed in the set of items to be analyzed; The traffic analysis unit 930 is used to perform traffic analysis on the first item to be analyzed based on a first cumulative exposure threshold and a preset conversion threshold, and obtain the analysis results. The potential identification unit 940 is used to determine the first item to be analyzed as the first item to be supported when the analysis result meets the potential identification conditions, and to perform potential identification on the first item to be supported to obtain the potential identification result. Support Unit 950 is used to provide traffic support to the first item to be supported when the potential identification result is high potential.

[0123] In some embodiments, the determining unit 920 is configured to: determine a first decision exposure threshold for the first item to be supported based on the current status information of the first item to be analyzed; determine the traffic analysis cycle and the remaining time of traffic analysis for the first item to be analyzed; and determine a first cumulative exposure threshold based on the traffic analysis cycle, the remaining time of traffic analysis, and the first decision exposure threshold.

[0124] In some embodiments, the traffic analysis unit 930 is configured to: acquire the cumulative exposure and cumulative conversion of the first item to be analyzed within a traffic analysis period; determine that the analysis result meets the potential identification conditions if the cumulative exposure is greater than or equal to a first cumulative exposure threshold and the cumulative conversion is greater than or equal to a preset conversion threshold, and / or the cumulative conversion is greater than or equal to the preset conversion threshold; and determine that the analysis result does not meet the potential identification conditions if the cumulative exposure is greater than or equal to the first cumulative exposure threshold and the cumulative conversion is less than the preset conversion threshold, and / or the cumulative conversion is less than the preset conversion threshold.

[0125] In some embodiments, the potential identification unit 940 is configured to: construct a first potential outcome model based on the item feature information, support score label, and individual treatment effect value to be estimated of the first item to be supported; train the first preset estimation model based on the item feature information and support score label to obtain a propensity score model; train the second preset estimation model based on the item feature information and the first potential outcome model to obtain an outcome prediction model; estimate the individual treatment effect value to be estimated based on the item feature information, support score label, propensity score model, and outcome prediction model to obtain an individual treatment effect value; and identify the potential of the first item to be supported based on the individual treatment effect value to obtain a potential identification result.

[0126] In some embodiments, the potential identification unit 940 is configured to: determine the effect estimation standard deviation corresponding to the individual treatment effect value; determine the potential identification result as high potential when the individual treatment effect value is greater than a preset treatment effect threshold and the effect estimation standard deviation is less than a preset effect estimation threshold; and determine the potential identification result as low potential when the individual treatment effect value is less than or equal to the preset treatment effect threshold, or the effect estimation standard deviation is greater than or equal to the preset effect estimation threshold.

[0127] In some embodiments, the support unit 950 is configured to: determine the first item to be supported as the first high-quality item when the potential identification result is high potential, and obtain a set of high-quality items including the first high-quality item; determine the treatment effect change rate of the first high-quality item based on the individual treatment effect value of the first high-quality item; obtain a set of target high-quality items based on the treatment effect change rate and a preset change rate threshold; and sort the at least one target high-quality item by score based on the individual treatment effect value of at least one target high-quality item in the set of target high-quality items, so as to provide traffic support to the at least one target high-quality item according to the score sorting result.

[0128] In some embodiments, the apparatus includes: an update unit, configured to determine a historical feedback dataset based on analysis results and potential identification results; update the current state information of a first item to be analyzed based on the historical feedback dataset to obtain target state information; and determine a first update cumulative exposure threshold associated with the first item to be analyzed based on the target state information, so as to perform traffic analysis on the first item to be analyzed according to the first update cumulative exposure threshold.

[0129] The methods and apparatus provided in the embodiments of this application have been described above. To implement the functions of the methods provided in the embodiments of this application, the electronic device may include a hardware structure and software modules, and may implement the above functions in the form of a hardware structure, software modules, or a hardware structure plus software modules. One of the above functions may be executed in the form of a hardware structure, software modules, or a hardware structure plus software modules.

[0130] Figure 10 This is a block diagram illustrating an electronic device 1000 for implementing the above-described traffic support method, according to an exemplary embodiment. For example, the electronic device 1000 may be a mobile phone, computer, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0131] Reference Figure 10 The electronic device 1000 may include a communication interface 1001, capable of interacting with other devices; a processor 1002, connected to the communication interface 1001 to interact with other devices, used to execute the methods provided by one or more of the above-described technical solutions when running a computer program; and a memory 1003, on which the computer program is stored. Specifically, the specific processing procedure of the processor 1002 can refer to the traffic support method described in the above embodiments of this application.

[0132] Of course, in practical applications, the various components in electronic device 1000 are coupled together through bus system 1004. It can be understood that bus system 1004 is used to realize the connection and communication between these components. In addition to a data bus, bus system 1004 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, in... Figure 10 The general labeled all buses as Bus System 1004.

[0133] The memory 1003 in this embodiment is used to store various types of data to support the operation of the electronic device 1000. Examples of such data include any computer program used to operate on the electronic device 1000.

[0134] The methods disclosed in the embodiments of this application can be applied to processor 1002, or implemented by processor 1002. Processor 1002 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 1002 or by instructions in the form of software. The processor 1002 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 1002 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. A general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in memory 1003. Processor 1002 reads the information in memory 1003 and combines its hardware to complete the steps of the aforementioned method.

[0135] In an exemplary embodiment, the electronic device 1000 may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0136] Embodiments of this application also propose a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the traffic support method described in the above embodiments of this application.

[0137] Embodiments of this application also propose a computer program product, including a computer program that is executed by a processor using the traffic support method described in the above embodiments of this application.

[0138] Embodiments of this application also propose a chip including one or more interface circuits and one or more processors; the interface circuits are used to receive signals from the memory of an electronic device and send signals to the processors, the signals including computer instructions stored in the memory, which, when executed by the processor, cause the electronic device to perform the traffic support method described in the above embodiments of this application.

[0139] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0140] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0141] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0142] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processing module, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (control method), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0143] It should be understood that various parts of the embodiments of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0144] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0145] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc.

[0146] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for providing traffic support, characterized in that, The method includes: Obtain a set of items to be analyzed, including at least one item to be analyzed; Based on the current status information of the first item to be analyzed, a first cumulative exposure threshold associated with the first item to be analyzed is determined, wherein the first item to be analyzed is any item to be analyzed in the set of items to be analyzed; Based on the first cumulative exposure threshold and the preset conversion threshold, traffic analysis is performed on the first item to be analyzed to obtain the analysis results; If the analysis results meet the potential identification conditions, the first item to be analyzed is determined to be the first item to be supported, and the potential of the first item to be supported is identified to obtain the potential identification result. If the potential identification result is high potential, traffic support will be provided to the first item to be supported.

2. The method according to claim 1, characterized in that, The determination of the first cumulative exposure threshold associated with the first item to be analyzed based on the current status information of the first item to be supported includes: Based on the current status information of the first item to be analyzed, a first decision exposure threshold for the first item to be supported is determined; Determine the flow analysis cycle and remaining flow analysis time for the first item to be analyzed; The first cumulative exposure threshold is determined based on the traffic analysis period, the remaining time of the traffic analysis, and the first decision exposure threshold.

3. The method according to claim 1, characterized in that, The traffic analysis of the first item to be analyzed, based on the first cumulative exposure threshold and the preset conversion threshold, yields the following analysis results: Obtain the cumulative exposure and cumulative conversion of the first item to be analyzed within the traffic analysis period; If the cumulative exposure is greater than or equal to the first cumulative exposure threshold and the cumulative conversion is greater than or equal to the preset conversion threshold, and / or the cumulative conversion is greater than or equal to the preset conversion threshold, the analysis result is determined to meet the potential identification condition. If the cumulative exposure is greater than or equal to the first cumulative exposure threshold and the cumulative conversion is less than the preset conversion threshold, and / or the cumulative conversion is less than the preset conversion threshold, the analysis result is determined to be that the potential identification condition is not met.

4. The method according to claim 1, characterized in that, The potential identification of the first item to be supported, and the resulting potential identification results, include: Based on the item characteristic information, support score label and individual treatment effect value to be estimated of the first item to be supported, a first potential outcome model is constructed. Based on the item feature information and the support score label, the first preset estimation model is trained to obtain the propensity score model. Based on the item feature information and the first potential result model, the second preset estimation model is trained to obtain the result prediction model; Based on the item feature information, the support score label, the propensity score model, and the outcome prediction model, the individual treatment effect value to be estimated is estimated to obtain the individual treatment effect value. Based on the individual treatment effect value, the potential of the first item to be supported is identified, and the potential identification result is obtained.

5. The method according to claim 4, characterized in that, The potential identification of the first item to be supported based on the individual treatment effect value, and the resulting potential identification results include: Determine the standard deviation of the effect estimate corresponding to the individual treatment effect value; If the individual treatment effect value is greater than a preset treatment effect threshold and the effect estimation standard deviation is less than a preset effect estimation threshold, the potential identification result is determined to be high potential. If the individual treatment effect value is less than or equal to a preset treatment effect threshold, or if the effect estimation standard deviation is greater than or equal to a preset effect estimation threshold, the potential identification result is determined to be low potential.

6. The method according to claim 4, characterized in that, The provision of traffic support to the first item to be supported when the potential identification result is high potential includes: If the potential identification result is high potential, the first item to be supported is determined as the first high-quality item, and a set of high-quality items including the first high-quality item is obtained. Based on the individual treatment effect value of the first premium item, determine the rate of change of the treatment effect of the first premium item; Based on the change rate of the treatment effect and the preset change rate threshold, a set of target high-quality items is obtained; Based on the individual treatment effect value of at least one target high-quality item in the target high-quality item set, the at least one target high-quality item is ranked by score, and traffic support is provided to the at least one target high-quality item according to the score ranking result.

7. The method according to claim 1, characterized in that, After identifying the potential of the first item to be supported and obtaining the potential identification result, the method further includes: Based on the analysis results and the potential identification results, the historical feedback dataset is determined; Based on the historical feedback dataset, the current status information of the first item to be analyzed is updated to obtain the target status information; Based on the target status information, a first updated cumulative exposure threshold associated with the first item to be analyzed is determined, so as to perform traffic analysis on the first item to be analyzed according to the first updated cumulative exposure threshold.

8. A flow support device, characterized in that, The device includes: The acquisition unit is used to acquire a set of items to be analyzed, including at least one item to be analyzed. The determining unit is used to determine a first cumulative exposure threshold associated with the first item to be analyzed based on the current state information of the first item to be analyzed, wherein the first item to be analyzed is any item to be analyzed in the set of items to be analyzed; The traffic analysis unit is used to perform traffic analysis on the first item to be analyzed based on the first cumulative exposure threshold and the preset conversion threshold, and obtain the analysis results; A potential identification unit is used to determine the first item to be analyzed as the first item to be supported when the analysis result meets the potential identification conditions, and to perform potential identification on the first item to be supported to obtain the potential identification result. The support unit is used to provide traffic support to the first item to be supported when the potential identification result is high potential.

9. An electronic device, characterized in that, include: The processor and the memory used to store computer programs that can run on the processor. When the processor is used to run the computer program, it performs the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.

Images