Data processing method and device, electronic equipment and computer storage medium

Abstract

The embodiment of the invention provides a data processing method and device, electronic equipment and a computer storage medium. The method comprises the steps of obtaining input data of a first historical time period; the input data comprises user data, advertisement related data and target commodity information; analyzing the input data of the first historical time period to obtain an analysis result, and obtaining delivery information obtained according to the analysis result; the delivery information comprises a to-be-delivered target commodity category and a target crowd scene; obtaining a target prediction model related to the delivery information, and predicting a predicted user sample by using the target prediction model to obtain a prediction result; the predicted user sample is obtained based on the input data of the first historical time period; and determining a directional crowd according to the prediction result.

Description

Technical Field

[0001] The present application relates to the field of e-commerce technology, and in particular to a data processing method, device, electronic device and computer storage medium. Background Art

[0002] With the continuous development of the e-commerce industry, e-commerce advertising has become a very important form of advertising and is well known to a large number of advertisers. For advertisers, how to set reasonable marketing goals and efficiently select target groups is a major demand. When advertisers place advertisements, they will use the data management platform (DMP) on the demand-side platform (DSP) to target the audience. The crowd targeting methods include but are not limited to: advertisers uploading offline crowd packages, customizing offline crowd packages based on rules or models, customizing real-time targeted crowd packages based on rules, or using other intelligent rule-based crowd targeting functions provided by the platform. However, the above-mentioned crowd targeting method using offline crowd packages has a single application scenario and is relatively limited; while the crowd targeting method using rules relies more on experience and has a higher threshold for use. Summary of the invention

[0003] The present application provides a data processing method, device, electronic device and computer storage medium.

[0004] The technical solution of this application is implemented as follows:

[0005] The present application provides a data processing method, the method comprising:

[0006] Acquire input data of a first historical time period; the input data includes user data, advertisement-related data, and target product information;

[0007] Analyze the input data of the first historical time period to obtain analysis results, and acquire delivery information obtained according to the analysis results; the delivery information includes target commodity categories and target population scenarios to be delivered;

[0008] Acquire a target prediction model related to the delivery information, and use the target prediction model to predict a predicted user sample to obtain a prediction result; the predicted user sample is obtained based on input data of the first historical time period;

[0009] According to the prediction results, a targeted population is determined.

[0010] The embodiment of the present application also proposes a data processing device, which includes an acquisition module, an analysis module, a prediction module and a determination module, wherein:

[0011] An acquisition module, used to acquire input data of a first historical time period; the input data includes user data, advertisement-related data, and target product information;

[0012] An analysis module, configured to analyze the input data of the first historical time period, obtain analysis results, and acquire delivery information obtained according to the analysis results; the delivery information includes target commodity categories and target population scenarios to be delivered;

[0013] A prediction module, used to obtain a target prediction model related to the delivery information, and use the target prediction model to predict a predicted user sample to obtain a prediction result; the predicted user sample is obtained based on the input data of the first historical time period;

[0014] A determination module is used to determine a targeted population based on the prediction result.

[0015] An embodiment of the present application provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the data processing method provided by one or more of the aforementioned technical solutions when executing the program.

[0016] An embodiment of the present application provides a computer storage medium, which stores a computer program; after the computer program is executed, it can implement the data processing method provided by one or more of the above-mentioned technical solutions.

[0017] The embodiment of the present application proposes a data processing method, device, electronic device and computer storage medium, the method comprising: obtaining input data of a first historical time period; the input data comprises user data, advertisement-related data and target product information; analyzing the input data of the first historical time period to obtain analysis results, and obtaining delivery information obtained according to the analysis results; the delivery information comprises target product categories and target population scenarios to be delivered; obtaining a target prediction model related to the delivery information, and using the target prediction model to predict a predicted user sample to obtain a prediction result; the predicted user sample is obtained based on the input data of the first historical time period; and determining a targeted population according to the prediction results.

[0018] It can be seen that the embodiment of the present application can obtain the target product category and target population scenario to be launched by analyzing the user data, advertising-related data and target product information, that is, it can establish a clear marketing goal; then, obtain a target prediction model related to the target product category and target population scenario to predict the user sample. If the determined target product category and target population scenario are different, the corresponding target prediction model may be different; in this way, targeted targeted population mining can be achieved for different population scenarios, thereby improving processing efficiency while ensuring the accuracy of the prediction results; thereby, improving the accuracy of population targeting. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1A is a flow chart of a data processing method in an embodiment of the present application;

[0020] Figure 1B It is a structural schematic diagram of a delivery system in an embodiment of the present application;

[0021] Figure 2 It is a schematic diagram of a process of training a CVR model in an embodiment of the present application;

[0022] Figure 3 is a structural schematic diagram of another delivery system according to an embodiment of the present application;

[0023] Figure 4 is a flowchart of another data processing method according to an embodiment of the present application;

[0024] Figure 5 is a schematic diagram of the structure of a data processing device according to an embodiment of the present application;

[0025] Figure 6 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the embodiments provided herein are only used to explain the present application and are not intended to limit the present application. In addition, the embodiments provided below are partial embodiments for implementing the present application, rather than providing all embodiments for implementing the present application. In the absence of conflict, the technical solutions recorded in the embodiments of the present application can be implemented in any combination.

[0027] It should be noted that, in the embodiments of the present application, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a method or apparatus including a series of elements includes not only the elements explicitly recorded, but also includes other elements not explicitly listed, or also includes elements inherent to the implementation of the method or apparatus. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other related elements (such as steps in a method or units in a device, for example, a unit may be a portion of a circuit, a portion of a processor, a portion of a program or software, etc.) in the method or apparatus including the element.

[0028] The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, I and / or J can represent the three situations that I exists alone, I and J exist at the same time, and J exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of I, J, and R can represent including any one or more elements selected from the set consisting of I, J, and R.

[0029] It should be noted that in the technical solution of the present invention, the collection, use, storage, sharing and transfer of user personal information involved are in compliance with the provisions of relevant laws and regulations, and it is necessary to inform the user and obtain the user's consent or authorization. When applicable, the user's personal information is de-identified and / or anonymized and / or encrypted.

[0030] Figure 1A is a flow chart of a data processing method in an embodiment of the present application, such as Figure 1A As shown, the method comprises the following steps:

[0031] Step 100: Obtain input data for a first historical time period.

[0032] In the embodiment of the present application, the data processing method can be applied to the delivery system, see Figure 1B The delivery system 20 may include a data input terminal 21, an insight analysis terminal 22, an algorithm processing terminal 23 and a delivery terminal 24. Figure 1B Each module shown illustrates the flow of the data processing method.

[0033] Exemplarily, the data input terminal can receive input data from multiple different data sources in the first historical time period; that is, the input data of the first historical time period can be obtained from the data input terminal; here, the input data can include user data, advertising-related data, and target product information.

[0034] Exemplarily, the first historical time period represents a historical time period of a preset length. Here, the length of the first historical time period can be set according to actual conditions, and the embodiment of the present application is not limited to this. For example, it can be a time period corresponding to the past 30 days, or the past 15 days, etc.; if it is assumed that the current day is the T day, the first historical time period can be from T-30 to T-1 (nearly 30 days), or from T-15 to T-1 (nearly 15 days).

[0035] Exemplarily, user data may include user behavior data and user portrait data on the e-commerce platform; wherein, user behavior data includes but is not limited to the user's browsing data, add-to-cart data, order data and collection data for any target product under the target product brand on the e-commerce platform.

[0036] Exemplarily, the delivery system can receive user data sent by the DMP platform through the data input terminal; wherein the data source of the DMP platform includes the e-commerce platform, that is, the DMP platform can obtain user data of the first historical time period from the e-commerce platform, and input it into the data input terminal of the delivery system after acquisition, so that the delivery system can obtain this part of data from the data input terminal.

[0037] Exemplarily, after the DMP platform obtains the user data of the first historical time period, this part of data can be referred to as user data within the DMP platform, referred to as in-site data, where in-site refers to within the DMP platform.

[0038] Exemplarily, the advertisement-related data may include data related to the advertisement provider, for example, it may be the one-party data of the advertisement provider; wherein the advertisement provider may be the advertiser itself or the advertiser's agent; the one-party data may be the advertiser's own data that is anonymized and then uploaded to the data input terminal of the delivery system, or it may be the advertiser's own data that is anonymized and then sent to the relevant agent, which is then uploaded to the data input terminal of the delivery system by the relevant agent.

[0039] For example, the first-party data includes but is not limited to member data, non-site channel order data, media platform exposure data, media platform click data, first-party tags, and customer relationship management (CRM) data, etc. Among them, each data in the first-party data has corresponding identification data for identification, which can be called the first-party data identification; when the advertising provider uploads the first-party data, it will also upload the first-party data identification to the data input terminal of the delivery system.

[0040] For example, the target product information is the product information that needs to be marketed and is provided by the advertisement provider; it may include only the target product brand, which may be expressed as <brand id>, or may include the target product brand and the product category under the brand, which may be expressed as <brand id, category id>, or may include partial product information under the target product brand category, which may be expressed as<sku_ids> The target product information can be uploaded by the advertisement provider to the data input terminal of the delivery system.

[0041] It should be noted that in addition to receiving the above-mentioned input data, the data input end can also receive real-time user data; among them, real-time user data refers to the return flow data generated by users' real-time interaction with the target product brand or product category inside and outside the site; here, inside and outside the site refers to inside and outside the DMP platform; real-time interactive behavior includes but is not limited to real-time browsing behavior, real-time add-to-cart behavior, and real-time collection behavior.

[0042] Step 101: Analyze input data of a first historical time period to obtain analysis results, and acquire delivery information obtained according to the analysis results.

[0043] In an embodiment of the present application, the input data of the first historical time period obtained can be analyzed through the insight analysis terminal to obtain analysis results, and delivery information obtained based on the analysis results can be obtained; here, the analysis results include recommended product categories and crowd scenarios; the delivery information includes target product categories and target crowd scenarios to be delivered.

[0044] Exemplarily, the target product category may be any category under the target product brand, which is not limited in the embodiments of the present application; for example, when the target product brand is a clothing brand, the target product category may be clothing, shoes, hats, etc.

[0045] Exemplarily, the target population scenario may be any population scenario of new products, hot products, new customers, old customers, first purchases, repeat purchases, gifts, high conversion populations, and promotion-sensitive populations, and the embodiments of the present application are not limited to this.

[0046] Exemplarily, the insight analysis end can analyze the input data in two different dimensions to obtain analysis results; one dimension is the brand dimension and the other dimension is the user dimension; specifically, the input data can be analyzed in the target product brand dimension to obtain the recommended product category, and further, based on the recommended product category, the input data can be analyzed in the user dimension to obtain the recommended population scenario.

[0047] Furthermore, after obtaining the above analysis results, the advertising provider can combine the analysis results with its own needs to obtain the delivery information; here, the delivery information includes the target product categories and target population scenarios to be delivered, that is, the target product categories and target population scenarios to be delivered by the advertising provider; after that, the delivery information can be sent to the algorithm processing end of the delivery system in the form of instructions, so that the algorithm processing end can obtain the delivery information. Here, the algorithm processing end mainly supports various population scenarios based on core algorithms and general capabilities. It can be seen that analyzing the input data through the insight analysis end can help the advertising provider further clarify its marketing goals.

[0048] Step 102: Obtain a target prediction model related to the delivery information, and use the target prediction model to predict the predicted user sample to obtain a prediction result.

[0049] In the embodiment of the present application, when the target population scenario is determined according to the above steps, the target prediction model related to the target population scenario is obtained through the algorithm processing end; illustratively, the target prediction model can be any of the following: conversion rate (Conversion Rate, CVR) model, exposure conversion rate (Click-Through & ConVersion Rate, CTCVR) model, similar population expansion (lookalike) model, behavior sequence modeling (behavior sequence) and gain model (Uplift Model). Among them, each target prediction model is used to mine the targeted population for the relevant target population scenario.

[0050] In some embodiments, the target prediction model is trained through the following steps: obtaining input data of the second historical time period, and obtaining a training user sample based on the input data of the second historical time period; performing feature processing on the input data of the second historical time period to obtain training feature data; and using the training user sample and the training feature data to train the initial prediction model to obtain the target prediction model.

[0051] Here, the second historical time period is the time period before the first historical time period, and the second historical time period may be the same length as the first historical time period or may be different; for example, assuming that the current day is the T day, the T-30th day to the T-15th day may be determined as the second historical time period, and the T-14th day to the T-1st day may be determined as the first historical time period; or the T-30th day to the T-16th day may be determined as the second historical time period, and the T-15th day to the T-1st day may be determined as the first historical time period.

[0052] It should be noted that the difference between the input data of the second historical time period and the input data of the first historical time period is only the time window, and the data type included in the input data is the same; here, the process of obtaining the input data of the second historical time period is not described in detail.

[0053] Exemplarily, based on the acquired advertisement-related data, user behavior data and user portrait data of the second historical time period, training user samples and training feature data can be generated, and an initial prediction model can be trained based on these two parts of data to obtain a target prediction model.

[0054] In the embodiment of the present application, the training user sample may include positive sample users and negative sample users, the positive sample users include users who have purchased the target product in the second historical time period, and the negative sample users include users who have not purchased the target product in the second historical time period. Here, the target product may be any product under the target product category.

[0055] Below, taking the target group scenario as a high conversion group as an example, Figure 2 The process of training the CVR model related to the high conversion population is exemplified. Here, there is no limitation on the implementation method of the CVR model, for example, it can be implemented by extreme gradient boosting (eXtreme Gradient Boosting, XGB) or light gradient boosting machine (Light Gradient Boosting Machine, LGB).

[0056] For example, by Figure 2 It can be seen that the data sources used by the CVR model in the modeling process include user portrait data of the second historical time period, user behavior data (including user browsing data, purchase data, order data and collection data on the site, etc.) and one-party data (corresponding to the above-mentioned advertising-related data); these data can be obtained from the input data of the second historical time period. After obtaining the above data, labeling processing is performed to obtain sample labels, that is, training user samples; the process of obtaining training user samples is exemplified below.

[0057] Exemplarily, when the second historical time period is from T-30 to T-16, taking T-30 as an example, the process of generating positive and negative user samples (i.e., positive user samples and negative user samples) for T-30 can be: select users who browse the target product category on T-30 and have target product purchase behavior on that day and the next 14 days (T-30 to T-16) as positive sample users. Among them, users who have multiple target product purchase behaviors on T-30 are counted as multiple positive sample users, and their purchase timestamps are used as unique identifiers; for user purchase behaviors other than T-30, they are counted as one positive sample user, so as to enhance the conversion of the day. Users who browse the target product category on T-30 and have no target product purchase behavior on that day and the next 14 days (T-30 to T-16) are selected as negative sample users. In this way, the positive and negative user samples from the remaining days T-29 to T-15 are determined respectively in the above manner, and the positive and negative user samples from the days T-30 to T-15 can be obtained. Finally, the set including these positive and negative user samples is determined as the training user sample.

[0058] Furthermore, after obtaining the training user sample, the training feature data can be obtained by performing feature processing on the party data, user behavior data and user portrait data in the second historical time period; Figure 2 It can be seen that the training feature data may include user portrait features, user behavior features, and one-party features; for example, user portrait features may include features such as the user's age, gender, and purchasing power; user behavior features may include features related to the user's browsing, adding to cart, placing orders, and collecting behaviors; and one-party features are obtained by processing the one-party data uploaded by the advertisement provider. The following is an explanation of the acquisition process of these three features.

[0059] For example, taking the T-30th day as an example, the user portrait features for the T-30th day can be obtained by: selecting the user portrait data of the user on the T-31st day as the feature; since the user portrait data has been labeled, it can be directly referenced as a feature; and so on, the user portrait features corresponding to the user portrait data of the second historical time period can be obtained.

[0060] Exemplarily, the number of times a user browses, adds to cart, places an order, and collects the target product under the target product category during the second historical time period can be used as a user behavior feature. It should be noted that the calculation logic of the user behavior features of the day is relatively special. User behavior features other than that of the day only need to be simply calculated in full. The user behavior features of the day need to consider whether the user behavior of the day occurs before or after the purchase of the target product. Only data that occurs before the purchase of the target product can be calculated into the user behavior features to reduce the risk of feature crossing and ensure the effectiveness of model training.

[0061] For example, the concept of user behavior characteristics for the day is explained as follows: for a positive sample user, if the positive sample user has a purchase behavior on the Tth day, the calculation of the user behavior characteristics for the day is based on the purchase timestamp; if the positive sample user has no purchase behavior on the Tth day, the calculation of the user behavior characteristics for the day is based on 24:00 on the day. For a negative sample user, the calculation of the user behavior characteristics for the day is based on 24:00 on the day.

[0062] Exemplarily, before uploading the first-party data to the delivery system, the advertising provider will anonymize the advertiser's own data (such as the first-party label). The processing process may be: anonymize the first-party label and moderately obfuscate the label value. This process will retain the order relationship between the data, bucket or one-hot encode the value, and then upload the anonymized and obfuscated label data to the delivery system; the first-party feature is obtained by processing the first-party data through outlier processing and feature screening.

[0063] Exemplarily, after obtaining training user samples and training feature data, the initial prediction model is trained using the training user samples and the training feature data to obtain a target prediction model, which may include: associating the training feature data with the training user samples to obtain a training sample set; dividing the training sample set according to a set ratio to obtain a training set and a test set; training the initial prediction model using the training set to obtain a trained initial prediction model; and adjusting parameters of the trained initial prediction model using the test set to obtain a target prediction model.

[0064] Exemplarily, by associating the user portrait features, user behavior features and one-party features included in the training feature data with each positive sample user in the training user sample, multiple positive samples can be obtained; similarly, by associating the user portrait features, user behavior features and one-party features included in the training feature data with each negative sample user in the training user sample, multiple negative samples can be obtained; and the set including multiple positive samples and multiple negative samples is determined as a training sample set.

[0065] Further, refer to Figure 2It can be seen that after obtaining the training sample set, sampling can be performed according to the set positive-negative sample ratio to obtain the sampled training sample set. During sampling, all positive samples are retained and only negative samples are downsampled. Here, the value of the positive-negative sample ratio can be determined according to the actual situation, and the embodiment of the present application does not limit this. For example, the value can be 1:10, etc.

[0066] For example, after obtaining the sampled training sample set, the training sample set is further divided according to the set ratio to obtain a training set and a test set; then, the training set and the test set are used to perform model training to obtain a target prediction model (corresponding to Figure 2 The target prediction model is related to the high conversion population. Here, the value of the set ratio can be determined according to the actual situation, and the embodiment of the present application does not limit this. For example, the value can be 4:1.

[0067] In an embodiment of the present application, after obtaining a target prediction model related to a target population scenario (e.g., a high conversion population), the above method further includes: determining whether the target prediction model is available; if it is determined that the target prediction model is available, determining a model threshold corresponding to the target prediction model.

[0068] Exemplarily, by simultaneously performing feature analysis and model analysis, the AUC (Area Under Curve) value of the target prediction model is determined, and then whether the target prediction model is available is determined based on the AUC value of the target prediction model; here, the correlation between the AUC value and whether the target prediction model is available can be determined based on actual conditions, and the embodiments of the present application are not limited to this. For example, if it is determined that the AUC of the target prediction model is greater than 0.8, then it is determined that the target prediction model is available; conversely, if it is determined that the AUC of the target prediction model is less than or equal to 0.8, then it is determined that the target prediction model is unavailable.

[0069] Furthermore, when it is determined that the target prediction model is available, the model threshold corresponding to the target prediction model is determined based on the value of the false positive rate (fpr); illustratively, the target prediction model can obtain a mapping relationship table of the false positive rate (fpr), the true positive rate (tpr), and the preset threshold based on the data of the test set; wherein, one fpr uniquely corresponds to one preset threshold, and the preset threshold when fpr=0.15 can be determined as the model threshold when the target prediction model is used online, and the embodiments of the present application are not limited to this.

[0070] In the embodiment of the present application, after completing the above steps, a predicted user sample is obtained, wherein the predicted user sample includes multiple predicted users, and the predicted user sample is obtained based on the input data of the first historical time period; it should be noted that the process of obtaining the predicted user sample is similar to the process of obtaining the training user sample, which will not be repeated here. Then, the target prediction model (for example, CVR model file) related to the target population scenario is used to predict the obtained predicted user sample.

[0071] Exemplarily, the process of predicting the acquired predicted user samples may include the following steps: feature processing the input data of the first historical time period to obtain predicted feature data; associating the predicted feature data with the predicted user samples to obtain a predicted sample set; then, predicting the predicted sample set using the target prediction model to obtain a prediction result; wherein the prediction result includes a predicted score for each predicted user. It should be noted that the process of determining the predicted feature data and the predicted sample set is similar to the process of determining the training feature data and the training sample set in the above training process, and will not be repeated here.

[0072] Step 103: Determine the targeted population based on the prediction results.

[0073] In an embodiment of the present application, after obtaining the prediction results and the model threshold corresponding to the target prediction model according to the above steps, the targeted population can be determined based on the prediction results and the model threshold corresponding to the target prediction model.

[0074] Exemplarily, the prediction score of each predicted user in the predicted user sample may be compared with the model threshold corresponding to the target prediction model, and each predicted user having a prediction score greater than the model threshold may be determined, and the set including these predicted users may be determined as the targeted population.

[0075] For example, assuming that the prediction score of a predicted user A is 0.8, and the model threshold corresponding to the target prediction model is 0.7, since the prediction score of the predicted user A is greater than the model threshold corresponding to the target prediction model, it means that the predicted user A matches the target product category. Therefore, the predicted user A will be added to the targeted population.

[0076] In some embodiments, the above method also includes: obtaining real-time user behavior data, and obtaining real-time user samples based on the real-time user behavior data; using the target prediction model to predict the real-time user samples to obtain the prediction results of the real-time user samples; obtaining real-time users according to the model threshold corresponding to the target prediction model and the prediction results of the real-time user samples; adding real-time users to the targeted population to obtain an updated targeted population.

[0077] Exemplarily, the targeted population is determined based on input data of a historical time period, which can be called an offline population; after obtaining the targeted population, the population can be brought online; wherein, the online population includes the online offline population and the online real-time stream model.

[0078] Exemplarily, bringing offline people online means pushing targeted people to the delivery end of the delivery system to achieve advertising delivery; the process may include: obtaining advertising information related to the target product category; selecting at least one delivery channel from the delivery channel list, and delivering the advertising information to the targeted people through at least one delivery channel.

[0079] Exemplarily, the delivery end pre-stores a delivery channel list, which includes but is not limited to search ads, touchpoint ads, direct ads, booth ads, media real-time interface (Real Time API, RTA) ads and other delivery channels.

[0080] Exemplarily, real-time user behavior data can be collected through real-time user behavior logs, and real-time user samples can be obtained based on the real-time user behavior data; it should be noted that the process of obtaining real-time user samples is similar to the process of obtaining training user samples mentioned above, and will not be repeated here.

[0081] Here, the user behavior data may include but is not limited to the time of occurrence of the user behavior, the category of the user behavior, the user's information, and the product information corresponding to the user behavior. The category of user behavior may include the user's browsing, collection, adding to cart, and purchase of a product.

[0082] Exemplarily, the real-time streaming model is put online by configuring the target prediction model into the crowd computing service; after the configuration is completed, the target prediction model will start processing the real-time user samples in the real-time streaming service, where the users involved in the real-time user samples can be browsing users, favorite users, added-to-cart users, or purchasing users. The target prediction model will predict the real-time user samples in the real-time streaming service and obtain the prediction results; the prediction results include the prediction scores of the real-time user samples, and by comparing the prediction scores of the real-time user samples with the model threshold corresponding to the target prediction model, it is determined whether the prediction scores of the real-time user samples are greater than the threshold. If it is determined to be, the real-time user samples are added to the targeted population as real-time users to obtain an updated targeted population. That is, the real-time user samples with high predicted conversion probability can be added to the targeted population in real time through the target prediction model to reach the real-time high conversion population.

[0083] Furthermore, the above method may also include: storing the real-time users after each real-time user is obtained; obtaining the real-time population stored within a preset time period; the real-time population includes at least one real-time user; determining a complementary population that does not have target commodity purchasing behavior from the real-time population; the complementary population includes at least one real-time user; adding the complementary population to the targeted population to obtain a targeted population that is updated again.

[0084] Here, the length of the prediction time period is not limited, for example, it can be nearly 1 day or nearly 3 days.

[0085] Exemplarily, each time a new real-time user is obtained, the real-time user can be stored offline. For the real-time population stored in the predicted time period, the users who have purchased products are removed, and only the complementary population needs to be continuously pushed to continuously reach the historical high-conversion population; here, the complementary population includes all real-time users in the real-time population who do not have any target product purchasing behavior.

[0086] It should be noted that each time a targeted population, an updated targeted population, or a targeted population that is updated again is obtained, advertising information related to the target product category can be obtained through the delivery end; at least one delivery channel is selected from the delivery channel list, and the advertising information is delivered to the above-mentioned population through at least one delivery channel.

[0087] The embodiment of the present application proposes a data processing method, which includes: obtaining input data of a first historical time period; the input data includes user data, advertising-related data and target product information; analyzing the input data of the first historical time period to obtain analysis results, and obtaining delivery information obtained according to the analysis results; the delivery information includes the target product category and target population scene to be delivered; obtaining a target prediction model related to the delivery information, and using the target prediction model to predict the predicted user sample to obtain a prediction result; the predicted user sample is obtained based on the input data of the first historical time period; according to the prediction result, determining the targeted population. It can be seen that the embodiment of the present application can obtain the target product category and target population scene to be delivered by analyzing user data, advertising-related data and target product information, that is, it can establish a clear marketing goal; then, obtain the target prediction model related to the target product category and target population scene to predict the predicted user sample, if the determined target product category and target population scene are different, then the corresponding target prediction model may be different; in this way, targeted targeted population mining can be achieved for different population scenes, improving processing efficiency while ensuring the accuracy of prediction results; thereby, improving the accuracy of population targeting.

[0088] In order to better reflect the purpose of this application, further explanation is given based on the above embodiments of this application.

[0089] Figure 3 is a structural diagram of another delivery system according to an embodiment of the present application. Figure 3 As shown, the data input terminal 21 includes party data, user data (corresponding to Figure 3 The insight analysis terminal 22 includes a brand analysis part and a user analysis part; the brand analysis part is used to realize brand sales diagnosis, query conversion analysis and category association analysis. The user analysis part is used to realize one-party user identification, global RFM (Recency Frenquency Moary) stratification, N-order repurchase analysis and swing user preference analysis.

[0090] Exemplarily, brand sales diagnosis is based on analyzing brand sales trends, product category sales trends, trends of several brands ranked at the top in the industry, and competitive brands of the top-selling product categories, to determine the recommended product categories. Query conversion analysis is based on analyzing the page views (PV), unique visitors (UV), order volume, and conversion rate of user advertising search terms and natural search terms, and is used to provide advertising providers with high-frequency natural search terms for non-advertising placements, and to provide guidance for advertisers to purchase search terms. Category association analysis analyzes users' cross-product category behavior, including cross-product category analysis within the same brand, cross-product category analysis on the entire e-commerce platform, and analysis of previous brand category browsing behavior, to help advertising providers select highly relevant product categories and conversion rates, and then formulate cross-product category strategies.

[0091] Exemplarily, in terms of user analysis, one-party user identification is used to match one-party users with e-commerce platform users, calculate the mapping rate, member matching rate, frequency distribution, portrait, 4A distribution, number of new platform customers and number of new brand customers, etc., to enhance the advertising provider's understanding of one-party users. Global RFM stratification combines the order data of the advertising provider and the e-commerce platform, and stratifies the old customers of the entire e-commerce platform based on the RFM model to help advertising providers understand the assets of old customers. N-order repurchase analysis is based on brand-purchased users, analyzing the user's minimum stock keeping unit (SKU, StockKeeping Unit) purchase link from the first order to multiple orders, and under the marketing demand of the first-order user to multiple-order users, it helps advertising providers select products and circle marketing groups for guidance. The swing user preference analysis is based on the SKU dimension to analyze the group of people who browse the target product and competing products within a certain period, and stratifies the swing group by frequency quantification, into those with high preference for target products or competing products and those with moderate preference for target products or competing products. At the same time, the conversion distribution of the swing group for target products and competing products is analyzed, which can help advertising providers diagnose the inflow and outflow of the swing group and identify real competing products.

[0092] Exemplarily, the algorithm processing end 23 mainly supports various crowd scenarios based on core algorithms and general capabilities; crowd scenarios include but are not limited to: new products, hot products, new customers, old customers, first purchases, repeat purchases, gifts, high-conversion crowds, and promotion-sensitive crowds. The core algorithm models used by the algorithm processing end 23 to support the above-mentioned crowd scenarios include CVR model, CTCVR model, lookalike model, behavior sequence modeling and gain model. After analysis by the insight analysis end, the advertising provider can establish clearer marketing goals based on its own needs, including what kind of products to market and promote, and based on which crowd scenarios to market. On this basis, based on general capabilities and core algorithm models, targeted crowds that meet various crowd scenarios can be provided to advertising providers.

[0093] Among them, the CVR model targets high-conversion populations for advertising providers based on the advertising provider's one-party data and user data. The CTCVR model takes into account the target optimization of click-through rate (CTR) and CVR, and takes into account the advertising provider's needs for mining click users. The lookalike model uses the one-party population provided by the advertising provider as the seed user, and the candidate population designated by the advertising provider as the candidate set to train the model, and uses the model to find similar populations to the seed population. Behavior sequence modeling uses the user's behavior sequence as a feature, and the conversion of the target product as the goal, to more widely mine people with conversion intentions. The gain model uses the user as the granularity to estimate the catalytic ability of advertising information on different users, which helps advertising providers to allocate marketing budgets in a refined manner.

[0094] The general capabilities of the algorithm processing end 23 include one-party feature engineering, model evaluation, traffic estimation, unified modeling, and crowd mining engine. One-party feature engineering converts one-party data uploaded by the advertising provider into feature data (corresponding to the above-mentioned one-party feature) through feature processing. Model evaluation judges the availability of the target prediction model based on the AUC value of the output target prediction model. Traffic estimation infers the amount of data processed by the target prediction model in a certain period of time (for example, daily) based on the online traffic situation and the configuration information of the target prediction model, which is used to control system resources. Unified modeling is different from single (bycase) modeling. It models the brand categories of the entire e-commerce platform and uses a unified model to mine the intended population of all brand categories. The crowd mining engine is used to mine the population of specified rules. The rules are generally obtained by insight analysis, and can also be specified by the advertising provider according to its own needs.

[0095] For example, see Figure 3The basic technology library used in the implementation process of the delivery system may include Hypertext Preprocessor (PHP), distributed computing engines Spark, Hadoop, Flink, Keras and Scikit-lean, etc. The main functional modules supporting the operation of the delivery system may include system background, offline data calculation, model training and real-time streaming services, etc. The delivery channels involved in the delivery end 24 may include search advertising, touch point advertising, direct advertising, booth advertising and media RTA advertising, etc.

[0096] It can be seen that the embodiments of the present application can combine user data and first-party data provided by the advertising provider to provide the advertising provider with a process framework from insight analysis to crowd targeting mining and multi-channel real-time delivery, which is used to help advertisers clarify their marketing goals while providing accurate crowd targeting services.

[0097] Figure 4 is a flow chart of another data processing method according to an embodiment of the present application. Figure 4 As shown, the process includes the following steps:

[0098] Step 30: Get input data.

[0099] Exemplarily, the input data of the advertising provider may include party data uploaded by the advertising provider, party data identifier, task type, target product information and other configuration information; here, the task type indicates the type of the target prediction model; other configuration information may be partial product information under the target product brand category.

[0100] Step 31: Determine the training sample set.

[0101] Exemplarily, the input data also includes acquired user data, and the algorithm processing end performs data calculation on the input data of the advertisement provider and the user data to obtain a training sample set; the algorithm processing end can also perform other data tasks to obtain other data results.

[0102] Step 32: Model training.

[0103] For example, after obtaining the training samples, the model training scheduling task is executed, that is, the model training is performed using the training samples, and the target prediction model (corresponding to Figure 4 Based on the model file, the predicted user sample can be predicted to obtain the targeted population (corresponding to Figure 4 By performing feature analysis and model analysis on the model file at the same time, it can be determined whether the target prediction model is available. If the target prediction model is available, the model threshold corresponding to the target prediction model can be obtained (corresponding to Figure 4Since this process has been described in the above embodiment, it will not be repeated here.

[0104] Step 33: The crowd goes online.

[0105] Exemplarily, it includes the online launch of offline crowds and the online launch of real-time streaming models; wherein, the process of the online launch of offline crowds can be: first, the offline crowds are used to register the crowds, and a unique crowd package is generated, which has a corresponding crowd package name and crowd package id. The real-time crowd refers to the crowd that is added to the crowd package in real time, and the supplementary crowd refers to the crowd that is added to the crowd package at preset time intervals.

[0106] For example, the real-time streaming model is launched by configuring the target prediction model, i.e., the model file, into the crowd computing service. After the configuration is completed, the target prediction model starts to process the real-time user samples in the real-time streaming service. The target prediction model predicts the real-time user samples in the real-time streaming service and obtains the prediction score of the real-time user samples (corresponding to Figure 4 Here, the crowd computing service will use the user's pin (corresponding to Figure 4 Request user service through user_pin) to obtain user portrait features and user behavior features corresponding to real-time user samples; request one-party feature online service through user_pin and one-party data identifier to obtain the user's one-party features; the target prediction model determines the prediction score of the real-time user sample based on the above feature data and real-time user samples, corresponding to Figure 4 (user_pin, P value). Further, the P value of the real-time user sample is compared with the model threshold. When it is determined that the P value is greater than the model threshold according to the comparison result, the real-time user sample is added to the targeted population as a real-time user. In addition, each time a real-time user is obtained, the real-time user can be stored offline. For the real-time population stored in the past k days (k is an integer greater than zero), the purchased users are removed to obtain the complement population, and the complement population is added to the targeted population.

[0107] It can be seen that the embodiment of the present application mines the users in the real-time population who have the intention to convert the target product through the target prediction model; and in the process of population targeting, the real-time users in the remarketing population with high model-predicted conversion probability are added to the targeted population in real time; for the real-time population saved in the recent time period, after removing the users who have purchased, the advertising information is pushed to the supplementary population who have not purchased, so as to achieve the purpose of continuously reaching the historical population that has not converted, and provide advertisers with more accurate population targeting services.

[0108] Figure 5 is a schematic diagram of the structure of the data processing device according to an embodiment of the present application. Figure 5As shown, the device includes: an acquisition module 400, an analysis module 401, a prediction module 402 and a determination module 403, wherein:

[0109] The acquisition module 400 is used to acquire input data of a first historical time period; the input data includes user data, advertisement-related data, and target product information;

[0110] The analysis module 401 is used to analyze the input data of the first historical time period, obtain analysis results, and acquire delivery information obtained according to the analysis results; the delivery information includes target commodity categories and target population scenarios to be delivered;

[0111] The prediction module 402 is used to obtain a target prediction model related to the delivery information, and use the target prediction model to predict the predicted user sample to obtain a prediction result; the predicted user sample is obtained based on the input data of the first historical time period;

[0112] The determination module 403 is used to determine the targeted population according to the prediction result.

[0113] In some embodiments, the apparatus further comprises a training module, wherein the training module is configured to:

[0114] Acquire input data of a second historical time period, and obtain a training user sample based on the input data of the second historical time period; the training user sample includes positive sample users and negative sample users, the positive sample users include users who have purchased the target product in the second historical time period, and the negative sample users include users who have not purchased the target product in the second historical time period; the second historical time period is a time period before the first historical time period;

[0115] Performing feature processing on the input data of the second historical time period to obtain training feature data;

[0116] The initial prediction model is trained using the training user samples and the training feature data to obtain the target prediction model.

[0117] In some embodiments, the training module is further used to:

[0118] Associating the training feature data with the training user samples to obtain a training sample set;

[0119] Dividing the training sample set according to a set ratio to obtain a training set and a test set;

[0120] Using the training set to train the initial prediction model to obtain a trained initial prediction model;

[0121] The test set is used to adjust the parameters of the trained initial prediction model to obtain the target prediction model.

[0122] In some embodiments, the determining module 403 is further configured to include:

[0123] Determining whether the target prediction model is available;

[0124] When it is determined that the target prediction model is available, determining a model threshold corresponding to the target prediction model;

[0125] The targeted population is determined according to the prediction result and the model threshold corresponding to the target prediction model.

[0126] In some embodiments, the apparatus further comprises an obtaining module, wherein the obtaining module is configured to:

[0127] Acquire real-time user data, and obtain real-time user samples based on the real-time user data;

[0128] Using the target prediction model to predict the real-time user sample, to obtain a prediction result of the real-time user sample;

[0129] Obtaining a real-time user according to a model threshold corresponding to the target prediction model and a prediction result of the real-time user sample;

[0130] The real-time users are added to the targeted population to obtain an updated targeted population.

[0131] In some embodiments, the obtaining module is further used to:

[0132] After obtaining the real-time user each time, storing the real-time user;

[0133] Acquire a real-time crowd stored within a preset time period; the real-time crowd includes at least one real-time user;

[0134] Determine a complementary group of people who do not have target commodity purchase behavior from the real-time group of people; the complementary group of people includes at least one real-time user;

[0135] The complement population is added to the targeted population to obtain an updated targeted population.

[0136] In some embodiments, the device further comprises a delivery module, wherein the delivery module is configured to:

[0137] Obtaining advertising information related to the target product category;

[0138] At least one delivery channel is selected from the delivery channel list, and the advertisement information is delivered to the targeted group of people through the at least one delivery channel; the delivery channel list includes a plurality of delivery channels.

[0139] In practical applications, the acquisition module 400, analysis module 401, prediction module 402, determination module 403, training module, obtaining module and delivery module can all be implemented by a processor located in an electronic device, and the processor can be at least one of an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, and a microprocessor.

[0140] In addition, each functional module in this embodiment can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or software functional modules.

[0141] If the integrated unit is implemented in the form of a software function module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment is essentially or the part that contributes to the relevant technology or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) or a processor to perform all or part of the steps of the method of this embodiment. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc. Various media that can store program codes.

[0142] Specifically, a computer program instruction corresponding to a data processing method in this embodiment can be stored on a storage medium such as a CD, a hard disk, or a USB flash drive. When the computer program instruction corresponding to a data processing method in the storage medium is read or executed by an electronic device, any data processing method in the aforementioned embodiments is implemented.

[0143] Based on the same technical concept as the above embodiments, see Figure 6 , which shows an electronic device 500 provided in an embodiment of the present application, which may include: a memory 501 and a processor 502; wherein,

[0144] Memory 501, used for storing computer programs and data;

[0145] The processor 502 is used to execute the computer program stored in the memory to implement any one of the data processing methods in the foregoing embodiments.

[0146] In practical applications, the memory 501 may be a volatile memory, such as RAM; or a non-volatile memory, such as ROM, flash memory, hard disk drive (HDD) or solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor 502.

[0147] The processor 502 may be at least one of ASIC, DSP, DSPD, PLD, FPGA, CPU, controller, microcontroller, and microprocessor. It is understandable that for different delivery platforms, the electronic device used to implement the processor function may also be other, which is not specifically limited in the embodiments of the present application.

[0148] In some embodiments, the functions or modules included in the device provided in the embodiments of the present application can be used to execute the method described in the method embodiment above, and its specific implementation can refer to the description of the method embodiment above. For the sake of brevity, it will not be repeated here. The above description of each embodiment tends to emphasize the differences between the various embodiments, and the same or similarities can refer to each other. For the sake of brevity, it will not be repeated here. The methods disclosed in the various method embodiments provided in this application can be arbitrarily combined without conflict to obtain a new method embodiment. The features disclosed in the various product embodiments provided in this application can be arbitrarily combined without conflict to obtain a new product embodiment. The features disclosed in the various method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain a new method embodiment or device embodiment.

[0149] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of hardware embodiments, software embodiments, or embodiments in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0150] The present application is described with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, and the combination of the process and / or box in the flowchart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one process or multiple processes in the flowchart and / or one box or multiple boxes in the block diagram.

[0151] These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0152] The above are only preferred embodiments of the present application and are not intended to limit the protection scope of the present application.

Claims

1. A data processing method, characterized in that: The method comprises: Acquire input data of a first historical time period; the input data includes user data, advertisement-related data, and target product information; Analyze the input data of the first historical time period to obtain analysis results, and acquire delivery information obtained according to the analysis results; the delivery information includes target commodity categories and target population scenarios to be delivered; Acquire a target prediction model related to the delivery information, and use the target prediction model to predict a predicted user sample to obtain a prediction result; the predicted user sample is obtained based on input data of the first historical time period; According to the prediction results, a targeted population is determined.

2. The method according to claim 1, characterized in that The target prediction model is trained by the following steps: Acquire input data of a second historical time period, and obtain a training user sample based on the input data of the second historical time period; the training user sample includes positive sample users and negative sample users, the positive sample users include users who have purchased the target product in the second historical time period, and the negative sample users include users who have not purchased the target product in the second historical time period; the second historical time period is a time period before the first historical time period; Performing feature processing on the input data of the second historical time period to obtain training feature data; The initial prediction model is trained using the training user samples and the training feature data to obtain the target prediction model.

3. The method according to claim 2, characterized in that The step of training the initial prediction model using the training user samples and the training feature data to obtain the target prediction model includes: Associating the training feature data with the training user samples to obtain a training sample set; Dividing the training sample set according to a set ratio to obtain a training set and a test set; Using the training set to train the initial prediction model to obtain a trained initial prediction model; The test set is used to adjust the parameters of the trained initial prediction model to obtain the target prediction model.

4. The method according to any one of claims 1 to 3, characterized in that: The method further comprises: Determining whether the target prediction model is available; When it is determined that the target prediction model is available, determining a model threshold corresponding to the target prediction model; Determining the targeted population according to the prediction results includes: The targeted population is determined according to the prediction result and the model threshold corresponding to the target prediction model.

5. The method according to claim 4, characterized in that The method further comprises: Acquire real-time user behavior data, and obtain real-time user samples based on the real-time user behavior data; Using the target prediction model to predict the real-time user sample, to obtain a prediction result of the real-time user sample; Obtaining a real-time user according to a model threshold corresponding to the target prediction model and a prediction result of the real-time user sample; The real-time users are added to the targeted population to obtain an updated targeted population.

6. The method according to claim 5, characterized in that The method further comprises: After obtaining the real-time user each time, storing the real-time user; Acquire a real-time crowd stored within a preset time period; the real-time crowd includes at least one real-time user; Determine a complementary group of people who do not have target commodity purchase behavior from the real-time group of people; the complementary group of people includes at least one real-time user; The complement population is added to the targeted population to obtain an updated targeted population.

7. The method according to any one of claims 1 to 3, characterized in that: The method further comprises: Obtaining advertising information related to the target product category; At least one delivery channel is selected from the delivery channel list, and the advertisement information is delivered to the targeted group of people through the at least one delivery channel; the delivery channel list includes a plurality of delivery channels.

8. A data processing device, characterized in that: The device comprises: An acquisition module, used to acquire input data of a first historical time period; the input data includes user data, advertisement-related data, and target product information; An analysis module, configured to analyze the input data of the first historical time period, obtain analysis results, and acquire delivery information obtained according to the analysis results; the delivery information includes target commodity categories and target population scenarios to be delivered; A prediction module, used to obtain a target prediction model related to the delivery information, and use the target prediction model to predict a predicted user sample to obtain a prediction result; the predicted user sample is obtained based on the input data of the first historical time period; A determination module is used to determine a targeted population based on the prediction result.

9. An electronic device, characterized in that: The device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the method according to any one of claims 1 to 7 is implemented when the processor executes the program.

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

Images