技术领域technical field
本发明实施例涉及数据处理技术领域，尤其涉及一种广告推荐方法、装置、终端和存储介质。The embodiments of the present invention relate to the technical field of data processing, and in particular to an advertisement recommendation method, device, terminal and storage medium.
背景技术Background technique
广告智能投放是个性化推荐系统的一大应用领域。运营广告投放的过程中，广告活动带来的点击率(Click Through Rate，CTR)和投资回报率(Return On Investment，ROI)是重要的评判标准。如何完善广告智能投放平台的推荐系统，从而将曝光量最大限度地有效化显得至关重要。Intelligent advertising delivery is a major application field of personalized recommendation system. In the process of operating advertisement delivery, the click through rate (Click Through Rate, CTR) and return on investment (Return On Investment, ROI) brought by the advertisement campaign are important criteria for judging. How to improve the recommendation system of the intelligent advertising platform to maximize the effectiveness of the exposure is very important.
传统广告智能投放平台的推荐过程往往面临两个方面的问题：个性化投放受众多因素如用户偏好、广告活动和用户额度等因素的限制；广告关联的商品类目多，且不同类目的商品产生的收益不均衡。针对上述问题，目前广告的推荐一般是根据用户历史消费或者点击等记录，总结用户偏好来匹配广告类目投放；基于历史数据统计曝光收益较高广告类目，给与相应类目的广告更多曝光机会，甚至是包站。但是上述广告推荐的方式存在以下缺陷：考虑用户单方面特征或者广告单方面特征，仅仅实现了针对用户或广告的个性化，准确性偏低；集中于投放收益高的广告类目，运营方式太过偏激且依赖人工配置，造成没有需求的用户对应的曝光浪费。The recommendation process of traditional advertising intelligent delivery platforms often faces two problems: personalized delivery is limited by many factors such as user preferences, advertising activities, and user quotas; there are many product categories associated with advertisements, and different categories of products The benefits generated are uneven. In view of the above problems, the current advertisement recommendation is generally based on the user's historical consumption or click records, summarizing user preferences to match the advertisement category; based on historical data statistics, the advertisement category with higher exposure income will give more advertisements to the corresponding category Exposure opportunities, even chartered stations. However, the above advertisement recommendation method has the following defects: Considering the unilateral characteristics of users or advertisements, it only realizes the personalization of users or advertisements, and the accuracy is low; It is too extreme and relies on manual configuration, resulting in waste of exposure for users who do not need it.
发明内容Contents of the invention
本发明实施例提供一种广告推荐方法、装置、终端和存储介质，以优化广告推荐方案，提高推荐的准确性，减少曝光浪费。Embodiments of the present invention provide an advertisement recommendation method, device, terminal, and storage medium to optimize an advertisement recommendation scheme, improve recommendation accuracy, and reduce exposure waste.
第一方面，本发明实施例提供了一种广告推荐方法，包括：In a first aspect, an embodiment of the present invention provides an advertisement recommendation method, including:
获取目标用户与待测广告的第一目标输入特征，以及所述目标用户与重点类目下所述待测广告的第二目标输入特征；Obtaining the first target input feature of the target user and the advertisement to be tested, and the second target input feature of the target user and the advertisement to be tested under the key category;
将所述第一目标输入特征输入预先构建的点击预测模型，得到所述目标用户对每个所述待测广告的点击概率；Inputting the first target input features into a pre-built click prediction model to obtain the click probability of each of the advertisements to be tested by the target user;
根据所述目标用户对每个所述待测广告的点击概率，确定所述待测广告的初始排序；determining the initial ranking of the advertisements to be tested according to the click probability of each advertisement to be tested by the target user;
将所述第二目标输入特征输入预先构建的下单预测模型，确定所述目标用户对重点类目下每个所述待测广告的下单概率；Inputting the second target input feature into a pre-built order prediction model to determine the probability of the target user placing an order for each of the advertisements to be tested under key categories;
根据所述目标用户对重点类目下每个所述待测广告的下单概率，对所述初始排序进行修正，确定所述待测广告的推荐排序。According to the target user's order probability for each of the advertisements to be tested under the key category, the initial ranking is corrected to determine the recommended ranking of the advertisements to be tested.
第二方面，本发明实施例还提供了一种广告推荐装置，该装置包括：In the second aspect, an embodiment of the present invention also provides an advertisement recommendation device, which includes:
特征获取模块，用于获取目标用户与待测广告的第一目标输入特征，以及所述目标用户与重点类目下所述待测广告的第二目标输入特征；A feature acquisition module, configured to acquire the first target input feature of the target user and the advertisement to be tested, and the second target input feature of the target user and the advertisement to be tested under the key category;
点击预测模块，用于将所述第一目标输入特征输入预先构建的点击预测模型，得到所述目标用户对每个所述待测广告的点击概率；A click prediction module, configured to input the first target input features into a pre-built click prediction model to obtain the click probability of each of the advertisements to be tested by the target user;
初始排序模块，用于根据所述目标用户对每个所述待测广告的点击概率，确定所述待测广告的初始排序；An initial ranking module, configured to determine the initial ranking of the advertisements to be tested according to the click probability of each of the advertisements to be tested by the target user;
下单预测模块，用于将所述第二目标输入特征输入预先构建的下单预测模型，确定所述目标用户对重点类目下每个所述待测广告的下单概率；An order prediction module, configured to input the second target input features into a pre-built order prediction model to determine the probability of the target user placing an order for each of the advertisements to be tested under key categories;
排序修正模块，用于根据所述目标用户对重点类目下每个所述待测广告的下单概率，对所述初始排序进行修正，确定所述待测广告的推荐排序。The order correction module is configured to correct the initial order according to the target user's order probability for each of the advertisements to be tested under key categories, and determine the recommended order of the advertisements to be tested.
进一步的，所述初始排序模块具体用于：Further, the initial sorting module is specifically used for:
根据所述目标用户对每个待测广告的点击概率，按照从大到小的顺序对所述待测广告进行排序，得到所述初始排序。According to the click probability of each advertisement to be tested by the target user, the advertisements to be tested are sorted in descending order to obtain the initial ranking.
进一步的，所述排序修正模块具体用于：Further, the sorting correction module is specifically used for:
对于属于所述重点类目，并且所述目标用户的下单概率小于概率阈值的所述待测广告，将其在所述初始排序中的位置按照设定规则进行打压，并修正所述初始排序。For the advertisements to be tested that belong to the key category and the order probability of the target user is less than the probability threshold, suppress their position in the initial ranking according to the set rules, and correct the initial ranking .
进一步的，该装置还包括：Further, the device also includes:
模型构建模块，用于在获取目标用户与待测广告的第一目标输入特征之前，构建所述点击预测模型和所述下单预测模型。The model construction module is used to construct the click prediction model and the order prediction model before obtaining the first target input features of the target user and the advertisement to be tested.
进一步的，所述模型构建模块包括点击预测模型单元，所述点击预测模型单元具体用于：Further, the model building module includes a click prediction model unit, and the click prediction model unit is specifically used for:
获取第一预设历史时间内的第一广告日志数据；Obtaining first advertisement log data within a first preset historical time;
根据所述第一广告日志数据生成第一样本数据，所述第一样本数据包括第一样本输入特征以及对应的点击标签，所述第一样本输入特征包括用户特征、广告特征以及用户与广告的交叉特征；Generate first sample data according to the first advertisement log data, the first sample data includes a first sample input feature and a corresponding click label, and the first sample input feature includes user features, advertisement features, and Cross-features of users and advertisements;
根据所述第一样本数据对高阶逻辑回归模型进行训练，得到点击预测模型。A high-order logistic regression model is trained according to the first sample data to obtain a click prediction model.
进一步的，所述模型构建模块包括下单预测模型单元，所述下单预测模型单元具体用于：Further, the model building module includes an order prediction model unit, and the order prediction model unit is specifically used for:
获取第二预设历史时间内，所述重点类目下的第二广告日志数据；Obtaining the second advertisement log data under the key category within the second preset historical time;
根据所述第二广告日志数据生成第二样本数据，所述第二样本数据包括第二样本输入特征以及对应的下单标签，所述第二样本输入特征包括用户特征、用户风控特征以及用户与所述重点类目下广告的交叉特征；Generate second sample data according to the second advertisement log data, the second sample data includes a second sample input feature and a corresponding order label, and the second sample input feature includes user features, user risk control features, and user Cross-features with advertisements under the key categories;
根据所述第二样本数据对Xgboost模型进行训练，得到下单预测模型。The Xgboost model is trained according to the second sample data to obtain an order prediction model.
进一步的，所述第一目标输入特征包括目标用户特征、待测广告特征以及所述目标用户与所述待测广告的交叉特征，所述第二目标输入特征包括目标用户特征、目标用户风控特征以及所述目标用户与所述重点类目下所述待测广告的交叉特征，所述目标用户为基于热点规则确定的用户，所述待测广告为设定目标时间内的有效广告。Further, the first target input feature includes the target user feature, the advertisement feature to be tested, and the intersection feature between the target user and the test advertisement, and the second target input feature includes the target user feature, the target user risk control features and the intersection features between the target user and the advertisement to be tested under the key category, the target user is a user determined based on hotspot rules, and the advertisement to be tested is an effective advertisement within a set target time.
第三方面，本发明实施例还提供了一种终端，所述终端包括：In a third aspect, an embodiment of the present invention further provides a terminal, where the terminal includes:
一个或多个处理器；one or more processors;
存储装置，用于存储一个或多个程序；storage means for storing one or more programs;
当所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现如上所述的广告推荐方法。When the one or more programs are executed by the one or more processors, the one or more processors implement the advertisement recommendation method as described above.
第四方面，本发明实施例还提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如上所述的广告推荐方法。In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the advertisement recommendation method as described above is implemented.
本发明实施例通过获取目标用户与待测广告的第一目标输入特征，以及目标用户与重点类目下待测广告的第二目标输入特征，将第一目标输入特征输入预先构建的点击预测模型，得到目标用户对每个待测广告的点击概率，并确定待测广告的初始排序；将第二目标输入特征输入预先构建的下单预测模型，确定目标用户对重点类目下每个待测广告的下单概率，并根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，确定待测广告的推荐排序。本发明实施例提供的技术方案，通过点击预测模型确定的点击概率确定待测广告的初始排序，并通过下单预测模型确定的下单概率对初始排序进行修正，最后得到待测广告的推荐排序，通过正向和反向的多角度考虑，给用户推荐最可能点击并且最可能下单的广告，可以在提高点击率的基础上降低无效曝光的概率，进而提升广告系统的整体曝光收益。In the embodiment of the present invention, by acquiring the first target input features of the target user and the advertisement to be tested, and the second target input features of the target user and the advertisement to be tested under key categories, input the first target input feature into the pre-built click prediction model , get the target user’s click probability for each advertisement to be tested, and determine the initial ranking of the advertisements to be tested; input the second target input feature into the pre-built order prediction model, and determine the target user’s click rate for each advertisement to be tested under the key category The order probability of the advertisement, and according to the order probability of the target user for each advertisement to be tested under the key category, the initial ranking is corrected to determine the recommended ranking of the advertisement to be tested. In the technical solution provided by the embodiment of the present invention, the initial ranking of the advertisements to be tested is determined by the click probability determined by the click prediction model, and the initial ranking is corrected by the order probability determined by the order prediction model, and finally the recommended ranking of the advertisements to be tested is obtained , through positive and negative multi-angle considerations, recommending advertisements that are most likely to be clicked and ordered by users can reduce the probability of invalid exposure on the basis of increasing the click-through rate, thereby increasing the overall exposure revenue of the advertising system.
附图说明Description of drawings
图1为本发明实施例一中的广告推荐方法的流程图；FIG. 1 is a flowchart of an advertisement recommendation method in Embodiment 1 of the present invention;
图2为本发明实施例一中的广告推荐方法的示意图；FIG. 2 is a schematic diagram of an advertisement recommendation method in Embodiment 1 of the present invention;
图3为本发明实施例一中的交叉特征的示意图；FIG. 3 is a schematic diagram of a cross feature in Embodiment 1 of the present invention;
图4为本发明实施例二中的广告推荐方法的流程图；FIG. 4 is a flowchart of an advertisement recommendation method in Embodiment 2 of the present invention;
图5为本发明实施例三中的广告推荐方法的流程图；FIG. 5 is a flowchart of an advertisement recommendation method in Embodiment 3 of the present invention;
图6为本发明实施例四中的广告推荐装置的结构示意图；FIG. 6 is a schematic structural diagram of an advertisement recommendation device in Embodiment 4 of the present invention;
图7为本发明实施例五中的终端的结构示意图。FIG. 7 is a schematic structural diagram of a terminal in Embodiment 5 of the present invention.
具体实施方式Detailed ways
下面结合附图和实施例对本发明作进一步的详细说明。可以理解的是，此处所描述的具体实施例仅仅用于解释本发明，而非对本发明的限定。另外还需要说明的是，为了便于描述，附图中仅示出了与本发明相关的部分而非全部结构。The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, but not to limit the present invention. In addition, it should be noted that, for the convenience of description, only some structures related to the present invention are shown in the drawings but not all structures.
实施例一Embodiment one
图1为本发明实施例一中的广告推荐方法的流程图，本实施例可适用于实现广告推荐的情况，该方法可以由广告推荐装置执行，该装置可以采用软件和/或硬件的方式实现，该装置可配置于终端中，例如该终端可以为智能手机、电脑和平板电脑等等。Figure 1 is a flow chart of the advertisement recommendation method in Embodiment 1 of the present invention. This embodiment is applicable to the situation of realizing advertisement recommendation. The method can be executed by an advertisement recommendation device, and the device can be realized by means of software and/or hardware. , the device can be configured in a terminal, for example, the terminal can be a smart phone, a computer, a tablet computer, and the like.
图2为本发明实施例一中的广告推荐方法的示意图，参见图2，广告系统可以包括广告后台11和广告前端12，其中广告前端12可以为用户的终端，本实施例中的广告推荐装置可以配置在广告前端12中，广告后台11可以为进行广告投放的后台，广告后台11中包括广告资源池，广告资源池中可以存储广告前端12需要展示给用户的全部广告资源和广告相关信息，例如展示时间等，广告资源池可以根据实际情况进行更新，例如定时更新。Fig. 2 is a schematic diagram of the advertisement recommendation method in Embodiment 1 of the present invention. Referring to Fig. 2, the advertisement system may include an advertisement background 11 and an advertisement front end 12, wherein the advertisement front end 12 may be a user's terminal, and the advertisement recommendation device in this embodiment It can be configured in the advertising front end 12, and the advertising background 11 can be the background for advertising. The advertising background 11 includes an advertising resource pool, which can store all the advertising resources and advertising related information that the advertising front end 12 needs to display to the user. For example, the display time, etc., the advertising resource pool can be updated according to the actual situation, such as regular update.
具体的，广告前端12可以进行在线日志拼接并将用户数据上报给服务器；当广告前端12中的广告推荐装置接收到广告推荐请求时，可以获取服务器中的用户数据，将通过特征工程处理之后的用户数据输入预先构建的模型中进行训练，得到训练好的模型，其中该模型可以根据实际情况进行设定，图中以LR模型(逻辑回归模型)为例；通过训练好的模型可以得到待展示广告的排序，即曝光点击率的排序表，根据该排序可以生成广告推荐列表；广告推荐装置可以将该广告推荐列表发送给广告前端12进行在线广告展示。其中，特征工程中可以包括用户基本信息、额度和还款信息、用户商品交互信息、广告基本信息、风控、购买意愿和广告内容特征等特征。可以理解的是，生成广告推荐列表之后，还可以根据相关规则进行过滤之后再发送给广告前端12。进一步的，图中广告前端12中的广告推荐装置还可以设置一个对照组，通过冷启动使得推荐的广告为随机的，进而对上述广告推荐列表的准确性进行验证。Specifically, the advertising front-end 12 can perform online log splicing and report user data to the server; when the advertising recommendation device in the advertising front-end 12 receives an advertisement recommendation request, it can obtain the user data in the server and process the user data through feature engineering. The user data is input into the pre-built model for training, and the trained model can be obtained. The model can be set according to the actual situation. In the figure, the LR model (logistic regression model) is taken as an example; the trained model can be obtained to be displayed The sorting of advertisements is the sorting table of exposure click-through rate, according to which an advertisement recommendation list can be generated; the advertisement recommendation device can send the advertisement recommendation list to the advertisement front end 12 for online advertisement display. Among them, feature engineering can include user basic information, credit line and repayment information, user product interaction information, basic advertising information, risk control, purchase intention, and advertising content characteristics. It can be understood that after the advertisement recommendation list is generated, it can also be filtered according to relevant rules before being sent to the advertisement front end 12 . Further, the advertisement recommendation device in the advertisement front end 12 in the figure can also set a control group, and make the recommended advertisements random by cold start, and then verify the accuracy of the above advertisement recommendation list.
如图1所示，该方法具体可以包括：As shown in Figure 1, the method may specifically include:
S110、获取目标用户与待测广告的第一目标输入特征，以及目标用户与重点类目下待测广告的第二目标输入特征。S110. Obtain the first target input features of the target user and the advertisement to be tested, and the second target input features of the target user and the advertisement to be tested under the key category.
其中，目标用户可以为基于热点规则确定的用户，该热点规则可以根据实际情况进行设定，例如目标用户可以为历史一个月内点击率大于点击率阈值的用户。待测广告可以为设定目标时间内的有效广告，其中设定目标时间可以根据实际情况进行设定，例如设定目标时间可以为一天、一周或一个月等，有效广告是指设定目标时间内与用户发生交互的广告，例如用户点击过的广告或用户分享过的广告等。目标用户和待测广告的数量本实施例中不作限定，可以根据实际情况进行设定。Wherein, the target user may be a user determined based on a hotspot rule, and the hotspot rule may be set according to actual conditions, for example, the target user may be a user whose click-through rate within a month is greater than a click-through rate threshold. The advertisement to be tested can be an effective advertisement within the set target time, wherein the set target time can be set according to the actual situation, for example, the set target time can be one day, one week or one month, etc., and an effective advertisement refers to the set target time Advertisements that interact with users in the website, such as advertisements clicked by users or advertisements shared by users. The number of target users and advertisements to be tested is not limited in this embodiment, and can be set according to actual conditions.
第一目标输入特征和第二目标输入特征为根据目标用户的相关数据和待测广告的相关数据，通过特征工程处理得到的特征。重点类目可以为当前广告投放平台通过历史数据分析，得到的具有集中投放的趋势或具备高覆盖率的广告类目，例如对于某个平台，其重点类目可以为3C类目，包括计算机(Computer)、通信(Communication)和消费类电子产品(Consumer Electronics)等。具体的，第一目标输入特征可以包括目标用户特征、待测广告特征以及目标用户与待测广告的交叉特征，第二目标输入特征可以包括目标用户特征、目标用户风控特征以及目标用户与重点类目下待测广告的交叉特征。The first target input feature and the second target input feature are features obtained through feature engineering processing based on the relevant data of the target user and the relevant data of the advertisement to be tested. The key categories can be the advertising categories with concentrated delivery or high coverage obtained through historical data analysis of the current advertising platform. For example, for a certain platform, the key categories can be 3C categories, including computer ( Computer), Communication (Communication) and Consumer Electronics (Consumer Electronics), etc. Specifically, the first target input features may include target user features, test advertisement features, and cross features between target users and test advertisements, and the second target input features may include target user features, target user risk control features, and target user and key features. The cross-features of the ads under test under the category.
目标用户特征可以包括待测用户的基本信息、风控及额度信息、历史偏好(如类目和品牌等)信息等，待测广告特征可以包括待测广告关联商品对应的类目和品牌等，目标用户和待测广告的交叉特征可以包括待测用户历史点击或购买待测广告匹配的类目和品牌等。具体的特征可以根据实际情况进行设定。The characteristics of the target user may include the basic information of the user to be tested, risk control and credit information, historical preferences (such as category and brand, etc.) information, etc. The characteristics of the advertisement to be tested may include the category and brand of the product associated with the advertisement to be tested, etc. The cross-features of the target user and the advertisement to be tested may include the category and brand matched by the user to be tested historically clicking or purchasing the advertisement to be tested. Specific features may be set according to actual conditions.
示例性的，图3为本发明实施例一中的交叉特征的示意图，图中的交叉特征为用户与3C类目中的广告的交叉特征，具体包括图中最近一个月3C曝光次数、最近一个月3C点击次数、最近一个月3C曝光点击率、最近三个月3C购买次数、最近一个月3C加车次数和最近三个月3C购买二审通过次数等等。Exemplarily, FIG. 3 is a schematic diagram of the cross feature in Embodiment 1 of the present invention. The cross feature in the figure is the cross feature between the user and the advertisement in the 3C category, specifically including the number of 3C exposures in the last month in the figure, the latest The number of 3C clicks per month, the click rate of 3C exposure in the last month, the number of 3C purchases in the last three months, the number of 3C car additions in the last month, the number of second-inspection passes for 3C purchases in the last three months, etc.
S120、将第一目标输入特征输入预先构建的点击预测模型，得到目标用户对每个待测广告的点击概率。S120. Input the first target input feature into the pre-built click prediction model to obtain the click probability of each advertisement to be tested by the target user.
其中，点击概率为用户点击广告的概率，可以通过百分数来表示，例如用户点击广告的概率可以为50％。获取到第一目标输入特征之后，可以将该第一目标输入特征输入预先构建的点击预测模型，输出目标用户对每个待测广告的点击概率。其中，该点击预测模型可以为基于深度学习模型进行训练得到，具体的深度学习模型本实施例中不作限定，例如点击预测模型可以基于高阶逻辑回归(Logistic Regression，LR)模型训练得到。Wherein, the click probability is the probability that the user clicks on the advertisement, which may be represented by a percentage, for example, the probability that the user clicks on the advertisement may be 50%. After the first target input feature is obtained, the first target input feature can be input into the pre-built click prediction model, and the target user's click probability for each advertisement to be tested can be output. Wherein, the click prediction model can be obtained by training based on a deep learning model, and the specific deep learning model is not limited in this embodiment, for example, the click prediction model can be obtained by training based on a high-order logistic regression (Logistic Regression, LR) model.
S130、根据目标用户对每个待测广告的点击概率，确定待测广告的初始排序。S130. Determine an initial ranking of the advertisements to be tested according to the click probability of each advertisement to be tested by the target user.
具体的，获取到目标用户对每个待测广告的点击概率之后，可以根据该点击概率，按照从大到小的顺序对待测广告进行排序，得到初始排序。该初始排序中，点击概率最大的待测广告排序最靠前，点击概率最小的待测广告排序最靠后。Specifically, after the target user's click probability for each advertisement to be tested is obtained, the advertisements to be tested may be sorted in descending order according to the click probability to obtain an initial ranking. In the initial sorting, the advertisement to be tested with the highest click probability is ranked first, and the advertisement to be tested with the lowest click probability is ranked last.
S140、将第二目标输入特征输入预先构建的下单预测模型，确定目标用户对重点类目下每个待测广告的下单概率。S140. Input the second target input feature into the pre-built order prediction model, and determine the target user's order probability for each advertisement to be tested under the key category.
在S110中获取到第二目标输入特征之后，可以将该第二目标输入特征输入下单预测模型中，输出目标用户对重点类目下每个待测广告的下单概率。其中，该下单预测模型可以为基于深度学习模型进行训练得到，具体的深度学习模型本实施例中不作限定，例如下单预测模型可以基于Xgboost模型训练得到。After the second target input feature is acquired in S110, the second target input feature can be input into the order prediction model, and the target user's order probability for each advertisement to be tested under the key category is output. Wherein, the order prediction model can be obtained by training based on a deep learning model, and the specific deep learning model is not limited in this embodiment, for example, the order prediction model can be obtained by training based on an Xgboost model.
S150、根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，确定待测广告的推荐排序。S150. Correct the initial ranking according to the target user's order probability for each advertisement to be tested under the key category, and determine the recommended ranking of the advertisements to be tested.
具体的，在S130中确定待测广告的初始排序的基础上，可以根据目标用户对重点类目下每个待测广告的下单概率，对该初始排序进行调整，将调整之后的初始排序确定为最终的推荐排序。Specifically, on the basis of determining the initial ranking of the advertisements to be tested in S130, the initial ranking can be adjusted according to the target user's order probability for each advertisement to be tested under the key category, and the adjusted initial ranking can be determined Rank the final recommendations.
可选地，根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，可以包括：对于属于重点类目，并且目标用户的下单概率小于概率阈值的待测广告，将其在初始排序中的位置按照设定规则进行打压，并修正初始排序。其中概率阈值和设定规则可以根据实际情况进行设定，例如概率阈值可以为0.5，设定规则可以为当前位置*10，本实施例中可以认为下单概率小于概率阈值的待测广告无推荐意义，在初始排序中的位置可以为打压到最后。示例性的，若在初始排序中待测广告m的位置在第2位，而该待测广告m属于重点类目并且目标用户对其的下单概率小于概率阈值，则将该待测广告m的位置通过当前位置*10的方式进行打压，得到待测广告m的位置调整完之后的初始排序。Optionally, according to the probability of the target user placing an order for each test advertisement under the key category, the initial ranking may be corrected, which may include: for the test results that belong to the key category and the target user's order probability is less than the probability threshold Advertisement, its position in the initial ranking is suppressed according to the set rules, and the initial ranking is corrected. The probability threshold and setting rules can be set according to the actual situation. For example, the probability threshold can be 0.5, and the setting rule can be current position * 10. In this embodiment, it can be considered that there is no recommendation for the advertisement to be tested whose order probability is less than the probability threshold Meaning, the position in the initial sorting can be suppressed to the end. Exemplarily, if the position of the advertisement m to be tested is in the second place in the initial ranking, and the advertisement m to be tested belongs to the key category and the probability of the target user placing an order for it is less than the probability threshold, then the advertisement m to be tested is The position of is suppressed by the method of current position * 10, and the initial ranking after the position adjustment of the advertisement m to be tested is obtained.
可选地，根据目标用户对于每个重点类目下的待预测广告的购买概率，对初始排序进行修正，还可以包括：根据目标用户对于每个属于重点类目的待测广告的当前可用额度，与历史成交价格阈值的比对结果，对初始排序进行修正。具体的，基于历史下单流水数据，可以得到目标用户下单重点类目中待测广告关联商品的成交价排序，可以将历史成交价格阈值设置为成交价格的90％，对比目标对于每个属于重点类目的待测广告的当前可用额度与历史成交价格阈值，对于当前可用额度小于历史成交价格阈值的待测广告，可以将其在初始排序中的位置按照设定规则进行打压。其中设定规则可以根据实际情况进行设定。Optionally, the initial ranking is modified according to the target user's purchase probability for each key category of the advertisement to be predicted, which may also include: according to the target user's current available quota for each of the key categories of the advertisement to be tested , the comparison result with the historical transaction price threshold, and the initial sorting is corrected. Specifically, based on the historical order flow data, the transaction price ranking of the advertising-related products to be tested in the key categories placed by the target user can be obtained, and the historical transaction price threshold can be set to 90% of the transaction price. The current available amount and historical transaction price threshold of the advertisements to be tested in key categories. For the advertisements to be tested whose current available amount is less than the historical transaction price threshold, their positions in the initial ranking can be suppressed according to the set rules. The setting rules can be set according to the actual situation.
通过对初始排序的修正，对于目标用户而言，通过待测广告的推荐排序投放的广告是基于用户个性化精准推荐，并且更可能产生订单收益的广告。Through the correction of the initial ranking, for the target users, the advertisements delivered through the recommended ranking of the advertisements to be tested are based on the user's personalized and accurate recommendation, and are more likely to generate order revenue.
本实施例通过获取目标用户与待测广告的第一目标输入特征，以及目标用户与重点类目下待测广告的第二目标输入特征，将第一目标输入特征输入预先构建的点击预测模型，得到目标用户对每个待测广告的点击概率，并确定待测广告的初始排序；将第二目标输入特征输入预先构建的下单预测模型，确定目标用户对重点类目下每个待测广告的下单概率，并根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，确定待测广告的推荐排序。本实施例提供的技术方案，通过点击预测模型确定的点击概率确定待测广告的初始排序，并通过下单预测模型确定的下单概率对初始排序进行修正，最后得到待测广告的推荐排序，通过正向和反向的多角度考虑，给用户推荐最可能点击并且最可能下单的广告，可以在提高点击通过率的基础上降低无效曝光的概率，进而提升广告系统的整体曝光收益。In this embodiment, by obtaining the first target input feature of the target user and the advertisement to be tested, and the second target input feature of the target user and the advertisement to be tested under the key category, the first target input feature is input into the pre-built click prediction model, Obtain the target user's click probability for each advertisement to be tested, and determine the initial ranking of the advertisements to be tested; input the second target input feature into the pre-built order prediction model, and determine the target user's click rate for each advertisement to be tested under the key category According to the order probability of the target user for each advertisement to be tested under the key category, the initial ranking is corrected to determine the recommended ranking of the advertisements to be tested. In the technical solution provided in this embodiment, the initial ranking of the advertisements to be tested is determined by the click probability determined by the click prediction model, and the initial ranking is corrected by the order probability determined by the order prediction model, and finally the recommended ranking of the advertisements to be tested is obtained. Through positive and negative multi-angle considerations, recommending advertisements that are most likely to be clicked and ordered by users can reduce the probability of invalid exposure on the basis of increasing the click-through rate, thereby increasing the overall exposure revenue of the advertising system.
实施例二Embodiment two
图4为本发明实施例二中的广告推荐方法的流程图。本实施例在上述实施例的基础上，进一步优化了上述广告推荐方法。相应的，如图4所示，本实施例的方法具体包括：FIG. 4 is a flow chart of the advertisement recommendation method in Embodiment 2 of the present invention. This embodiment further optimizes the above advertisement recommendation method on the basis of the above embodiments. Correspondingly, as shown in Figure 4, the method in this embodiment specifically includes:
S210、构建点击预测模型和下单预测模型。S210. Construct a click prediction model and an order prediction model.
具体的，构建点击预测模型，可以包括：获取第一预设历史时间内的第一广告日志数据；根据第一广告日志数据生成第一样本数据，第一样本数据包括第一样本输入特征以及对应的点击标签，第一样本输入特征包括用户特征、广告特征以及用户与广告的交叉特征；根据第一样本数据对高阶逻辑回归模型进行训练，得到点击预测模型。Specifically, constructing the click prediction model may include: obtaining first advertisement log data within a first preset historical time; generating first sample data according to the first advertisement log data, the first sample data including the first sample input Features and corresponding click labels, the first sample input features include user features, advertisement features, and cross features between users and advertisements; the high-order logistic regression model is trained according to the first sample data to obtain a click prediction model.
其中，第一预设历史时间可以根据实际情况进行设定，本实施例中以第一预设历史时间为一周为例进行说明。第一广告日志数据为与广告相关的数据，可以包括用户数据、广告数据和用户与广告的交互数据等数据。第一样本数据为点击预测模型的样本数据，第一样本数据中包括用于训练模型的训练样本数据和用于测试模型的测试样本数据。第一样本数据中包括多个样本，具体样本的数量本实施例中不作限定。每个样本可以包括第一样本输入特征及该第一样本输入特征对应的点击标签，该点击标签用于标记用户在第一预设历史时间内是否点击过对应的广告，若点击过，则点击标签标记为1，若未点击，则点击标签标记为-1。例如，若用户A对广告A在一周内的点击次数为10，该用户A对应的样本中的第一样本输入特征包括用户A的特征、广告A的特征和用户A与广告A的交叉特征，点击标签为1。Wherein, the first preset historical time can be set according to the actual situation. In this embodiment, the first preset historical time is one week as an example for illustration. The first advertisement log data is data related to advertisements, and may include data such as user data, advertisement data, and interaction data between users and advertisements. The first sample data is sample data of the click prediction model, and the first sample data includes training sample data for training the model and test sample data for testing the model. The first sample data includes multiple samples, and the specific number of samples is not limited in this embodiment. Each sample may include a first sample input feature and a click tag corresponding to the first sample input feature, the click tag is used to mark whether the user has clicked on the corresponding advertisement within the first preset historical time, and if so, The clicked label is marked as 1, and if not clicked, the clicked label is marked as -1. For example, if user A clicks 10 times on advertisement A within a week, the first sample input features in the sample corresponding to user A include the features of user A, the features of advertisement A, and the intersection features of user A and advertisement A , and the click label is 1.
构建下单预测模型，可以包括：获取第二预设历史时间内，重点类目下的第二广告日志数据；根据第二广告日志数据生成第二样本数据，第二样本数据包括第二样本输入特征以及对应的下单标签，第二样本输入特征包括用户特征、用户风控特征以及用户与重点类目下广告的交叉特征；根据第二样本数据对Xgboost模型进行训练，得到下单预测模型。其中，用户风控特征可以为表示对用户金融层面风险控制的特征，用户风控特征可以包括当月待还本金、用户待还本金、已还费用、当月应还本金、总信用额度、可用消费额度、已用消费额度、剩余免息额度和授信天数等等。Constructing an order prediction model may include: acquiring second advertisement log data under key categories within a second preset historical period; generating second sample data according to the second advertisement log data, the second sample data including the second sample input Features and corresponding order tags, the input features of the second sample include user features, user risk control features, and the intersection features of users and advertisements under key categories; the Xgboost model is trained according to the second sample data to obtain an order prediction model. Among them, the user's risk control feature can be a feature that represents the risk control of the user's financial level. The user's risk control feature can include the principal to be repaid in the current month, the principal to be repaid by the user, the repaid fee, the principal to be repaid in the current month, the total credit limit, Available consumption quota, used consumption quota, remaining interest-free quota and credit days, etc.
其中，第二预设历史时间可以根据实际情况进行设定，本实施例中以第二预设历史时间为一个月为例进行说明。第二广告日志数据为重点类目下与广告相关的数据，可以包括用户数据、广告数据和用户与广告的交互数据等数据。第二样本数据为点击预测模型的样本数据，第二样本数据中包括用于训练模型的训练样本数据和用于测试模型的测试样本数据。第二样本数据中包括多个样本，具体样本的数量本实施例中不作限定。每个样本可以包括第二样本输入特征及该第一样本输入特征对应的下单标签，该下单标签用于标记用户在第二预设历史时间内是否下单过对应广告的商品，若下单过，则下单标签标记为1，若未点击，则下单标签标记为-1。例如，若用户B对重点类目下的广告B对应的商品在一个月内的购买次数为5，该用户B对应的样本中的第二样本输入特征包括用户B的特征、用户B的风控特征和用户B与广告B的交叉特征，下单标签为1。Wherein, the second preset historical time can be set according to the actual situation. In this embodiment, the second preset historical time is one month as an example for illustration. The second advertisement log data is data related to advertisements under key categories, and may include data such as user data, advertisement data, and interaction data between users and advertisements. The second sample data is sample data of the click prediction model, and the second sample data includes training sample data for training the model and test sample data for testing the model. The second sample data includes multiple samples, and the specific number of samples is not limited in this embodiment. Each sample may include a second sample input feature and an order tag corresponding to the first sample input feature, and the order tag is used to mark whether the user has placed an order for the product corresponding to the advertisement within the second preset historical time, if If the order has been placed, the order label is marked as 1, and if not clicked, the order label is marked as -1. For example, if user B purchases 5 products corresponding to advertisement B under key categories within one month, the second sample input features in the sample corresponding to user B include user B's features, user B's risk control feature and the intersection feature of user B and ad B, the order label is 1.
进一步的，根据第一样本数据对高阶逻辑回归模型进行训练时以及根据第二样本数据对Xgboost模型进行训练时，均可以通过AUC(Area Under the Curve)和KS(Kolmogorov Smirnov)评价指标等指标衡量模型的可用性，若可用，则得到点击预测模型和下单预测模型。其中，AUC是指ROC曲线(Receiver Operating Characteristic curve)下的面积，是深度学习中的一种模型评估指标。Further, when training the high-order logistic regression model based on the first sample data and when training the Xgboost model based on the second sample data, both can pass AUC (Area Under the Curve) and KS (Kolmogorov Smirnov) evaluation indicators, etc. The indicator measures the usability of the model, and if available, the click prediction model and order prediction model are obtained. Among them, AUC refers to the area under the ROC curve (Receiver Operating Characteristic curve), which is a model evaluation index in deep learning.
S220、获取目标用户与待测广告的第一目标输入特征，以及目标用户与重点类目下待测广告的第二目标输入特征。S220. Obtain the first target input features of the target user and the advertisement to be tested, and the second target input features of the target user and the advertisement to be tested under the key category.
其中，第一目标输入特征包括目标用户特征、待测广告特征以及目标用户与待测广告的交叉特征，第二目标输入特征包括目标用户特征、目标用户风控特征、待测广告特征以及目标用户与重点类目下待测广告的交叉特征，目标用户为基于热点规则确定的用户，待测广告为设定目标时间内的有效广告。Wherein, the first target input feature includes the target user feature, the advertisement feature to be tested, and the intersection feature of the target user and the advertisement to be tested, and the second target input feature includes the target user feature, the target user risk control feature, the test advertisement feature, and the target user feature. The cross-features with the advertisements to be tested under the key categories, the target users are users determined based on the hotspot rules, and the advertisements to be tested are effective advertisements within the set target time.
S230、将第一目标输入特征输入预先构建的点击预测模型，得到目标用户对每个待测广告的点击概率。S230. Input the first target input feature into the pre-built click prediction model to obtain the click probability of each advertisement to be tested by the target user.
S240、根据目标用户对每个待测广告的点击概率，确定待测广告的初始排序。S240. Determine an initial ranking of the advertisements to be tested according to the click probability of each advertisement to be tested by the target user.
具体的，获取到目标用户对每个待测广告的点击概率之后，可以根据该点击概率，按照从大到小的顺序对待测广告进行排序，得到初始排序。Specifically, after the target user's click probability for each advertisement to be tested is obtained, the advertisements to be tested may be sorted in descending order according to the click probability to obtain an initial ranking.
S250、将第二目标输入特征输入预先构建的下单预测模型，确定目标用户对重点类目下每个待测广告的下单概率。S250. Input the second target input feature into the pre-built order prediction model, and determine the target user's order probability for each advertisement to be tested under the key category.
S260、根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，确定待测广告的推荐排序。S260. Correct the initial ranking according to the target user's order probability for each advertisement to be tested under the key category, and determine the recommended ranking of the advertisements to be tested.
具体的，在S240中确定待测广告的初始排序的基础上，可以根据目标用户对重点类目下每个待测广告的下单概率，对该初始排序进行调整，将调整之后的初始排序确定为最终的推荐排序。Specifically, on the basis of determining the initial ranking of the advertisements to be tested in S240, the initial ranking can be adjusted according to the target user's order probability for each advertisement to be tested under the key category, and the adjusted initial ranking can be determined Rank the final recommendations.
可选地，根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，可以包括：对于属于重点类目，并且目标用户的下单概率小于概率阈值的待测广告，将其在初始排序中的位置按照设定规则进行打压，并修正初始排序。Optionally, according to the probability of the target user placing an order for each test advertisement under the key category, the initial ranking may be corrected, which may include: for the test results that belong to the key category and the target user's order probability is less than the probability threshold Advertisement, its position in the initial ranking is suppressed according to the set rules, and the initial ranking is corrected.
S270、根据待测广告的推荐排序进行广告推荐。S270. Perform advertisement recommendation according to the recommendation ranking of the advertisements to be tested.
具体的，对于目标用户，确定待测广告最终的推荐排序之后，可以根据该推荐排序，推荐设定数量的待测广告给目标用户。其中，设定数量可以根据实际情况进行设定，例如设定数量可以为5或10等。Specifically, for the target user, after determining the final recommended ranking of the advertisements to be tested, a set number of advertisements to be tested may be recommended to the target user according to the recommended ranking. Wherein, the set number can be set according to the actual situation, for example, the set number can be 5 or 10 and so on.
本实施例构建点击预测模型和下单预测模型，通过获取目标用户与待测广告的第一目标输入特征，以及目标用户与重点类目下待测广告的第二目标输入特征，将第一目标输入特征输入预先构建的点击预测模型，得到目标用户对每个待测广告的点击概率，并确定待测广告的初始排序；将第二目标输入特征输入预先构建的下单预测模型，确定目标用户对重点类目下每个待测广告的下单概率，并根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，确定待测广告的推荐排序。本实施例提供的技术方案，通过点击预测模型确定的点击概率确定待测广告的初始排序，并通过下单预测模型确定的下单概率对初始排序进行修正，最后得到待测广告的推荐排序，通过正向和反向的多角度考虑，给用户推荐最可能点击并且最可能下单的广告，可以在提高点击通过率的基础上降低无效曝光的概率，进而提升广告系统的整体曝光收益；并且，通过用户与广告的交叉特征训练点击预测模型和下单预测模型，进一步提高了模型的准确性。This embodiment constructs a click prediction model and an order prediction model. By obtaining the first target input features of the target user and the advertisement to be tested, and the second target input features of the target user and the advertisement to be tested under key categories, the first target Input features into the pre-built click prediction model to obtain the target user's click probability for each advertisement to be tested, and determine the initial ranking of the advertisements to be tested; input the second target input feature into the pre-built order prediction model to determine the target user According to the order probability of each advertisement to be tested under the key category, and according to the probability of the target user placing an order for each advertisement to be tested under the key category, the initial ranking is corrected to determine the recommended ranking of the advertisement to be tested. In the technical solution provided in this embodiment, the initial ranking of the advertisements to be tested is determined by the click probability determined by the click prediction model, and the initial ranking is corrected by the order probability determined by the order prediction model, and finally the recommended ranking of the advertisements to be tested is obtained. Through positive and negative multi-angle considerations, recommending advertisements that are most likely to be clicked and placed by users can reduce the probability of invalid exposure on the basis of improving the click-through rate, thereby increasing the overall exposure revenue of the advertising system; and , the click prediction model and the order prediction model are trained through the cross-features of users and advertisements, which further improves the accuracy of the model.
实施例三Embodiment three
图5为本发明实施例三中的广告推荐方法的流程图。本实施例可以上述实施例为基础，通过一个具体的示例对广告推荐方法进行说明。本实施例中第一预设历史时间以一周为例进行说明，第二预设历史时间以一个月为例进行说明。参见图5，该方法具体可以包括：FIG. 5 is a flow chart of the advertisement recommendation method in Embodiment 3 of the present invention. This embodiment can be based on the foregoing embodiments, and a specific example is used to describe an advertisement recommendation method. In this embodiment, the first preset historical time is described by taking one week as an example, and the second preset historical time is described by taking one month as an example. Referring to Figure 5, the method specifically may include:
S301、开始。S301, start.
S302、近一周内曝光和点击数据，通过特征工程处理。S302. The exposure and click data in the past week is processed through feature engineering.
具体的，提取历史最近一周内的曝光和点击数据，即第一广告日志数据，通过特征工程处理，得到第一样本数据，将第一样本数据分为图5中的训练集和测试集。Specifically, extract the exposure and click data within the last week of history, that is, the first advertisement log data, and obtain the first sample data through feature engineering processing, and divide the first sample data into the training set and the test set in Figure 5 .
S303、匹配当天有效广告。S303 , matching the effective advertisement of the current day.
根据当天的实时数据确定当天的有效广告，即确定待测广告。The effective advertisement of the day is determined according to the real-time data of the day, that is, the advertisement to be tested is determined.
S304、初始化高阶LR模型。S304. Initialize the high-order LR model.
S305、训练高阶LR模型。S305. Train a high-order LR model.
将第一样本数据中的训练集输入高阶逻辑回归(Logistic Regression，LR)模型中，进行训练。Input the training set in the first sample data into a high-order logistic regression (Logistic Regression, LR) model for training.
S306、测试高阶LR模型。S306. Test the high-order LR model.
通过第一样本数据中的测试集对训练好的高阶LR模型进行测试。The trained high-order LR model is tested by the test set in the first sample data.
S307、AUC是否可接受。S307, whether AUC is acceptable.
判断高阶LR模型的AUC(Area Under the Curve)是否满足要求，若满足要求，则得到点击预测模型，执行S308，若不满足要求，则返回执行S302。Determine whether the AUC (Area Under the Curve) of the high-order LR model meets the requirements. If the requirements are met, the click prediction model is obtained, and S308 is executed. If the requirements are not met, the execution is returned to S302.
S308、预测目标用户对当天有效广告的点击概率。S308. Predict the click probability of the target user on the effective advertisement of the day.
将目标用户与当天有效广告的相关数据通过特征工程处理后的第一目标输入特征，输入测试完成的高阶LR模型，可以得到目标用户对当天有效广告的点击概率。The relevant data of the target user and the effective advertisement of the day is input into the first target feature after feature engineering processing, and input into the high-level LR model after the test, and the click probability of the target user on the effective advertisement of the day can be obtained.
S309、初始排序。S309. Initial sorting.
根据目标用户对当天有效广告的点击概率，按照从大到小的顺序对当天有效广告进行排序，得到初始排序。According to the target user's click probability on the effective advertisements of the day, the effective advertisements of the day are sorted in descending order, and the initial ranking is obtained.
S310、确定广告平台偏好类目。S310. Determine the advertising platform preference category.
确定当前广告平台的广告偏好类目，偏好类目的数量本实施例中不作限定。The advertisement preference categories of the current advertisement platform are determined, and the number of preference categories is not limited in this embodiment.
S311、近一个月内目标用户与偏好类目下广告的交互数据，通过特征工程处理。S311. The interaction data between the target user and the advertisement under the preference category in the past month is processed through feature engineering.
提取近一个月内目标用户与偏好类目下广告的交互数据，即第二广告日志数据，通过特征工程处理，得到第二样本数据，将第二样本数据分为图5中的训练集和测试集。Extract the interaction data between the target user and the advertisement under the preference category in the past month, that is, the second advertisement log data, and obtain the second sample data through feature engineering processing, and divide the second sample data into the training set and the test set in Figure 5 set.
S312、匹配当天有效广告。S312. Matching the effective advertisement of the current day.
根据当天的实时数据确定广告平台偏好类目下当天的有效广告，即确定广告平台偏好类目下的待测广告。According to the real-time data of the day, an effective advertisement of the day under the preference category of the advertising platform is determined, that is, an advertisement to be tested under the preference category of the advertising platform is determined.
S313、初始化XGB模型。S313. Initialize the XGB model.
S314、训练XGB模型。S314. Train the XGB model.
将第二样本数据中的训练集输入Xgboost(XGB)模型中，进行训练。Input the training set in the second sample data into the Xgboost (XGB) model for training.
S315、测试XGB模型。S315. Test the XGB model.
通过第二样本数据中的测试集对训练好的XGB模型进行测试。Test the trained XGB model through the test set in the second sample data.
S316、AUC是否可接受。S316, whether AUC is acceptable.
判断XGB模型的AUC(Area Under the Curve)是否满足要求，若满足要求，则得到下单预测模型，执行S317，若不满足要求，则返回执行S311。Judging whether the AUC (Area Under the Curve) of the XGB model meets the requirements, if the requirements are met, the order prediction model is obtained, and S317 is executed; if the requirements are not met, the execution returns to S311.
S317、预测目标用户对偏好类目下当天有效广告的下单概率。S317. Predict the probability of the target user placing an order for the current day's valid advertisement under the preferred category.
将目标用户与偏好类目下当天有效广告的相关数据通过特征工程处理后的第二目标输入特征，输入测试完成的XGB模型，可以得到目标用户对偏好类目下当天有效广告的下单概率。The target user and the relevant data of the current day's effective advertisement under the preferred category are input into the second target feature after feature engineering processing, and input into the tested XGB model, and the target user's order probability for the current day's effective advertisement under the preferred category can be obtained.
S318、推荐排序。S318 , recommend sorting.
根据目标用户对偏好类目下当天有效广告的下单概率对S309中确定的当天有效广告的初始排序进行调整，得到最终的推荐排序。The initial ranking of the current day's valid advertisements determined in S309 is adjusted according to the target user's order probability of the current day's valid advertisements under the preferred category to obtain a final recommendation ranking.
S319、相关业务规则过滤。S319. Filtering by relevant business rules.
确定推荐排序之后，还可以通过其他相关的业务规则对推荐排序进行过滤，例如通过部门ID去重对推荐排序进行过滤。After the recommendation ranking is determined, the recommendation ranking can also be filtered through other related business rules, for example, the recommendation ranking can be filtered by deduplication of department IDs.
S320、传递广告前端。S320, delivering the advertisement front end.
将过滤后的当天有效广告的推荐排序发送给广告前端，以展示给用户。Send the recommended order of the filtered effective advertisements of the day to the front end of the advertisement for display to the user.
S321、结束。S321, end.
本实施例中通过一个具体的示例对广告推荐方法进行了进一步地说明，本实施例通过点击预测模型确定的点击概率确定待测广告的初始排序，并通过下单预测模型确定的下单概率对初始排序进行修正，最后得到待测广告的推荐排序，通过正向和反向的多角度考虑，给用户推荐最可能点击并且最可能下单的广告，可以在提高点击通过率的基础上降低无效曝光的概率，进而提升广告系统的整体曝光收益。In this embodiment, a specific example is used to further illustrate the advertisement recommendation method. In this embodiment, the initial ranking of the advertisements to be tested is determined by the click probability determined by the click prediction model, and the order probability determined by the order prediction model is The initial ranking is corrected, and finally the recommended ranking of the advertisements to be tested is obtained. Through forward and reverse multi-angle considerations, the advertisements that are most likely to be clicked and most likely to be placed are recommended to users, which can reduce invalidity on the basis of increasing the click-through rate. Exposure probability, thereby increasing the overall exposure revenue of the advertising system.
实施例四Embodiment Four
图6为本发明实施例四中的广告推荐装置的结构示意图，本实施例可适用于实现广告推荐的情况。本发明实施例所提供的广告推荐装置可执行本发明任意实施例所提供的广告推荐方法，具备执行方法相应的功能模块和有益效果。FIG. 6 is a schematic structural diagram of an advertisement recommendation device in Embodiment 4 of the present invention. This embodiment is applicable to the case of implementing advertisement recommendation. The advertisement recommendation device provided in the embodiment of the present invention can execute the advertisement recommendation method provided in any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method.
该装置具体包括特征获取模块410、点击预测模块420、初始排序模块430、下单预测模块440和排序修正模块450，其中：The device specifically includes a feature acquisition module 410, a click prediction module 420, an initial ranking module 430, an order prediction module 440 and a ranking correction module 450, wherein:
特征获取模块410，用于获取目标用户与待测广告的第一目标输入特征，以及目标用户与重点类目下待测广告的第二目标输入特征；A feature acquisition module 410, configured to acquire the first target input feature of the target user and the advertisement to be tested, and the second target input feature of the target user and the advertisement to be tested under the key category;
点击预测模块420，用于将第一目标输入特征输入预先构建的点击预测模型，得到目标用户对每个待测广告的点击概率；The click prediction module 420 is used to input the first target input feature into the pre-built click prediction model to obtain the click probability of each advertisement to be tested by the target user;
初始排序模块430，用于根据目标用户对每个待测广告的点击概率，确定待测广告的初始排序；The initial ranking module 430 is used to determine the initial ranking of the advertisements to be tested according to the click probability of each advertisement to be tested by the target user;
下单预测模块440，用于将第二目标输入特征输入预先构建的下单预测模型，确定目标用户对重点类目下每个待测广告的下单概率；The order prediction module 440 is used to input the second target input feature into the pre-built order prediction model to determine the order probability of the target user for each advertisement to be tested under the key category;
排序修正模块450，用于根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，确定待测广告的推荐排序。The ranking correction module 450 is configured to correct the initial ranking according to the target user's order probability for each advertisement to be tested under the key category, and determine the recommended ranking of the advertisement to be tested.
本发明实施例构建点击预测模型和下单预测模型，通过获取目标用户与待测广告的第一目标输入特征，以及目标用户与重点类目下待测广告的第二目标输入特征，将第一目标输入特征输入预先构建的点击预测模型，得到目标用户对每个待测广告的点击概率，并确定待测广告的初始排序；将第二目标输入特征输入预先构建的下单预测模型，确定目标用户对重点类目下每个待测广告的下单概率，并根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，确定待测广告的推荐排序。本发明实施例提供的技术方案，通过点击预测模型确定的点击概率确定待测广告的初始排序，并通过下单预测模型确定的下单概率对初始排序进行修正，最后得到待测广告的推荐排序，通过正向和反向的多角度考虑，给用户推荐最可能点击并且最可能下单的广告，可以在提高点击通过率的基础上降低无效曝光的概率，进而提升广告系统的整体曝光收益。The embodiment of the present invention constructs a click prediction model and an order prediction model. By obtaining the first target input feature of the target user and the advertisement to be tested, and the second target input feature of the target user and the advertisement to be tested under the key category, the first Input the target input features into the pre-built click prediction model to obtain the target user’s click probability for each advertisement to be tested, and determine the initial ranking of the ads to be tested; input the second target input features into the pre-built order prediction model to determine the target The probability of the user placing an order for each advertisement to be tested under the key category, and according to the probability of the target user placing an order for each advertisement to be tested under the key category, the initial ranking is corrected to determine the recommended ranking of the advertisement to be tested. In the technical solution provided by the embodiment of the present invention, the initial ranking of the advertisements to be tested is determined by the click probability determined by the click prediction model, and the initial ranking is corrected by the order probability determined by the order prediction model, and finally the recommended ranking of the advertisements to be tested is obtained , through positive and negative multi-angle considerations, recommending advertisements that are most likely to be clicked and ordered by users can reduce the probability of invalid exposure on the basis of increasing the click-through rate, thereby increasing the overall exposure revenue of the advertising system.
进一步的，初始排序模块430具体用于：Further, the initial sorting module 430 is specifically used for:
根据目标用户对每个待测广告的点击概率，按照从大到小的顺序对待测广告进行排序，得到初始排序。According to the target user's click probability on each advertisement to be tested, the advertisements to be tested are sorted in descending order to obtain the initial ranking.
进一步的，排序修正模块450具体用于：Further, the sorting correction module 450 is specifically used for:
对于属于重点类目，并且目标用户的下单概率小于概率阈值的待测广告，将其在初始排序中的位置按照设定规则进行打压，并修正初始排序。For the advertisements to be tested that belong to the key categories and the order probability of the target user is less than the probability threshold, their position in the initial ranking is suppressed according to the set rules, and the initial ranking is corrected.
进一步的，该装置还包括：Further, the device also includes:
模型构建模块，用于在获取目标用户与待测广告的第一目标输入特征之前，构建点击预测模型和下单预测模型。The model construction module is used to construct a click prediction model and an order prediction model before obtaining the first target input features of the target user and the advertisement to be tested.
进一步的，模型构建模块包括点击预测模型单元，点击预测模型单元具体用于：Further, the model building module includes a click prediction model unit, and the click prediction model unit is specifically used for:
获取第一预设历史时间内的第一广告日志数据；Obtaining first advertisement log data within a first preset historical time;
根据第一广告日志数据生成第一样本数据，第一样本数据包括第一样本输入特征以及对应的点击标签，第一样本输入特征包括用户特征、广告特征以及用户与广告的交叉特征；Generate the first sample data according to the first advertisement log data, the first sample data includes the first sample input feature and the corresponding click label, the first sample input feature includes the user feature, the advertisement feature and the intersection feature of the user and the advertisement ;
根据第一样本数据对高阶逻辑回归模型进行训练，得到点击预测模型。The high-order logistic regression model is trained according to the first sample data to obtain a click prediction model.
进一步的，模型构建模块包括下单预测模型单元，下单预测模型单元具体用于：Further, the model building module includes an order prediction model unit, which is specifically used for:
获取第二预设历史时间内，重点类目下的第二广告日志数据；Obtain the second advertisement log data under key categories within the second preset historical time;
根据第二广告日志数据生成第二样本数据，第二样本数据包括第二样本输入特征以及对应的下单标签，第二样本输入特征包括用户特征、用户风控特征以及用户与重点类目下广告的交叉特征；Generate the second sample data according to the second advertisement log data, the second sample data includes the second sample input features and the corresponding order label, the second sample input features include user features, user risk control features, and users and advertisements under key categories cross features;
根据第二样本数据对Xgboost模型进行训练，得到下单预测模型。The Xgboost model is trained according to the second sample data to obtain an order prediction model.
进一步的，第一目标输入特征包括目标用户特征、待测广告特征以及目标用户与待测广告的交叉特征，第二目标输入特征包括目标用户特征、目标用户风控特征以及目标用户与重点类目下待测广告的交叉特征，目标用户为基于热点规则确定的用户，待测广告为设定目标时间内的有效广告。Further, the first target input feature includes the target user feature, the advertisement feature to be tested, and the cross feature between the target user and the test advertisement, and the second target input feature includes the target user feature, the target user risk control feature, and the target user and key category Under the cross-features of the advertisement to be tested, the target user is the user determined based on the hotspot rule, and the advertisement to be tested is an effective advertisement within the set target time.
本发明实施例所提供的广告推荐装置可执行本发明任意实施例所提供的广告推荐方法，具备执行方法相应的功能模块和有益效果。The advertisement recommendation device provided in the embodiment of the present invention can execute the advertisement recommendation method provided in any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method.
实施例五Embodiment five
图7为本发明实施例五中的终端的结构示意图。图7示出了适于用来实现本发明实施方式的示例性终端512的框图。图7显示的终端512仅仅是一个示例，不应对本发明实施例的功能和使用范围带来任何限制。FIG. 7 is a schematic structural diagram of a terminal in Embodiment 5 of the present invention. Figure 7 shows a block diagram of an exemplary terminal 512 suitable for use in implementing embodiments of the present invention. The terminal 512 shown in FIG. 7 is only an example, and should not limit the functions and scope of use of this embodiment of the present invention.
如图7所示，终端512以通用终端的形式表现。终端512的组件可以包括但不限于：一个或者多个处理器516，存储装置528，连接不同系统组件(包括存储装置528和处理器516)的总线518。As shown in FIG. 7, terminal 512 takes the form of a general-purpose terminal. Components of the terminal 512 may include, but are not limited to: one or more processors 516, a storage device 528, and a bus 518 connecting different system components (including the storage device 528 and the processor 516).
总线518表示几类总线结构中的一种或多种，包括存储装置总线或者存储装置控制器，外围总线，图形加速端口，处理器或者使用多种总线结构中的任意总线结构的局域总线。举例来说，这些体系结构包括但不限于工业标准体系结构(Industry SubversiveAlliance，ISA)总线，微通道体系结构(Micro Channel Architecture，MAC)总线，增强型ISA总线、视频电子标准协会(Video Electronics Standards Association，VESA)局域总线以及外围组件互连(Peripheral Component Interconnect，PCI)总线。Bus 518 represents one or more of several types of bus structures, including a storage device bus or controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (Industry Subversive Alliance, ISA) bus, Micro Channel Architecture (Micro Channel Architecture, MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (Video Electronics Standards Association , VESA) local bus and peripheral component interconnect (Peripheral Component Interconnect, PCI) bus.
终端512典型地包括多种计算机系统可读介质。这些介质可以是任何能够被终端512访问的可用介质，包括易失性和非易失性介质，可移动的和不可移动的介质。Terminal 512 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by terminal 512 and include both volatile and nonvolatile media, removable and non-removable media.
存储装置528可以包括易失性存储器形式的计算机系统可读介质，例如随机存取存储器(Random Access Memory，RAM)530和/或高速缓存存储器532。终端512可以进一步包括其它可移动/不可移动的、易失性/非易失性计算机系统存储介质。仅作为举例，存储系统534可以用于读写不可移动的、非易失性磁介质(图7未显示，通常称为“硬盘驱动器”)。尽管图7中未示出，可以提供用于对可移动非易失性磁盘(例如“软盘”)读写的磁盘驱动器，以及对可移动非易失性光盘，例如只读光盘(Compact Disc Read-Only Memory，CD-ROM),数字视盘(Digital Video Disc-Read Only Memory，DVD-ROM)或者其它光介质)读写的光盘驱动器。在这些情况下，每个驱动器可以通过一个或者多个数据介质接口与总线518相连。存储装置528可以包括至少一个程序产品，该程序产品具有一组(例如至少一个)程序模块，这些程序模块被配置以执行本发明各实施例的功能。The storage device 528 may include computer system readable media in the form of volatile memory, such as random access memory (Random Access Memory, RAM) 530 and/or cache memory 532 . Terminal 512 may further include other removable/non-removable, volatile/nonvolatile computer system storage media. By way of example only, storage system 534 may be used to read and write to non-removable, non-volatile magnetic media (not shown in FIG. 7, commonly referred to as a "hard drive"). Although not shown in FIG. 7, a disk drive for reading and writing to a removable nonvolatile disk (such as a "floppy disk") may be provided, as well as a removable nonvolatile disk, such as a Compact Disc Read Only Disk (Compact Disc Read-Only). -Only Memory, CD-ROM), Digital Video Disc (Digital Video Disc-Read Only Memory, DVD-ROM) or other optical media) CD-ROM drive. In these cases, each drive may be connected to bus 518 through one or more data media interfaces. Storage device 528 may include at least one program product having a set (eg, at least one) of program modules configured to perform the functions of various embodiments of the present invention.
具有一组(至少一个)程序模块542的程序/实用工具540，可以存储在例如存储装置528中，这样的程序模块542包括但不限于操作系统、一个或者多个应用程序、其它程序模块以及程序数据，这些示例中的每一个或某种组合中可能包括网络环境的实现。程序模块542通常执行本发明所描述的实施例中的功能和/或方法。A program/utility tool 540 having a set (at least one) of program modules 542, such as but not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include the implementation of the network environment. Program modules 542 generally perform the functions and/or methodologies of the described embodiments of the invention.
终端512也可以与一个或多个外部设备514(例如键盘、指向终端、显示器524等)通信，还可与一个或者多个使得用户能与该终端512交互的终端通信，和/或与使得该终端512能与一个或多个其它计算终端进行通信的任何终端(例如网卡，调制解调器等等)通信。这种通信可以通过输入/输出(I/O)接口522进行。并且，终端512还可以通过网络适配器520与一个或者多个网络(例如局域网(Local Area Network，LAN)，广域网(Wide Area Network，WAN)和/或公共网络，例如因特网)通信。如图7所示，网络适配器520通过总线518与终端512的其它模块通信。应当明白，尽管图中未示出，可以结合终端512使用其它硬件和/或软件模块，包括但不限于：微代码、终端驱动器、冗余处理器、外部磁盘驱动阵列、磁盘阵列(Redundant Arrays of Independent Disks，RAID)系统、磁带驱动器以及数据备份存储系统等。The terminal 512 may also communicate with one or more external devices 514 (such as a keyboard, a pointing terminal, a display 524, etc.), and may also communicate with one or more terminals that enable a user to interact with the terminal 512, and/or communicate with the Terminal 512 is any terminal capable of communicating with one or more other computing terminals (eg, a network card, modem, etc.). Such communication may occur through input/output (I/O) interface 522 . Moreover, the terminal 512 can also communicate with one or more networks (such as a local area network (Local Area Network, LAN), a wide area network (Wide Area Network, WAN) and/or a public network such as the Internet) through the network adapter 520. As shown in FIG. 7 , network adapter 520 communicates with other modules of terminal 512 via bus 518 . It should be appreciated that although not shown, other hardware and/or software modules may be used in conjunction with terminal 512, including but not limited to: microcode, terminal drivers, redundant processors, external disk drive arrays, disk arrays (Redundant Arrays of Independent Disks, RAID) systems, tape drives, and data backup storage systems.
处理器516通过运行存储在存储装置528中的程序，从而执行各种功能应用以及数据处理，例如实现本发明实施例所提供的广告推荐方法，该方法包括：The processor 516 executes various functional applications and data processing by running the program stored in the storage device 528, such as realizing the advertisement recommendation method provided by the embodiment of the present invention, the method includes:
获取目标用户与待测广告的第一目标输入特征，以及目标用户与重点类目下待测广告的第二目标输入特征；Obtain the first target input feature of the target user and the advertisement to be tested, and the second target input feature of the target user and the advertisement to be tested under the key category;
将第一目标输入特征输入预先构建的点击预测模型，得到目标用户对每个待测广告的点击概率；Inputting the first target input feature into the pre-built click prediction model to obtain the click probability of each advertisement to be tested by the target user;
根据目标用户对每个待测广告的点击概率，确定待测广告的初始排序；Determine the initial ranking of the advertisements to be tested according to the click probability of each advertisement to be tested by the target user;
将第二目标输入特征输入预先构建的下单预测模型，确定目标用户对重点类目下每个待测广告的下单概率；Input the second target input feature into the pre-built order prediction model to determine the probability of the target user placing an order for each advertisement to be tested under the key category;
根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，确定待测广告的推荐排序。According to the probability of the target user placing an order for each advertisement to be tested under the key category, the initial ranking is corrected to determine the recommended ranking of the advertisement to be tested.
实施例六Embodiment six
本发明实施例六还提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如本发明实施例所提供的广告推荐方法，该方法包括：Embodiment 6 of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the advertisement recommendation method provided in the embodiment of the present invention is implemented. The method includes:
获取目标用户与待测广告的第一目标输入特征，以及目标用户与重点类目下待测广告的第二目标输入特征；Obtain the first target input feature of the target user and the advertisement to be tested, and the second target input feature of the target user and the advertisement to be tested under the key category;
将第一目标输入特征输入预先构建的点击预测模型，得到目标用户对每个待测广告的点击概率；Inputting the first target input feature into the pre-built click prediction model to obtain the click probability of each advertisement to be tested by the target user;
根据目标用户对每个待测广告的点击概率，确定待测广告的初始排序；Determine the initial ranking of the advertisements to be tested according to the click probability of each advertisement to be tested by the target user;
将第二目标输入特征输入预先构建的下单预测模型，确定目标用户对重点类目下每个待测广告的下单概率；Input the second target input feature into the pre-built order prediction model to determine the probability of the target user placing an order for each advertisement to be tested under the key category;
根据目标用户对重点类目下每个待测广告的下单概率，对初始排序进行修正，确定待测广告的推荐排序。According to the probability of the target user placing an order for each advertisement to be tested under the key category, the initial ranking is corrected to determine the recommended ranking of the advertisement to be tested.
本发明实施例的计算机存储介质，可以采用一个或多个计算机可读的介质的任意组合。计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质。计算机可读存储介质例如可以是——但不限于——电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本文件中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。The computer storage medium in the embodiments of the present invention may use any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: electrical connections with one or more leads, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), Erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
计算机可读的信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。计算机可读的信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读介质可以发送、传播或者传输用于由指令执行系统、装置或者器件使用或者与其结合使用的程序。A computer readable signal medium may include a data signal carrying computer readable program code in baseband or as part of a carrier wave. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. .
计算机可读介质上包含的程序代码可以用任何适当的介质传输，包括——但不限于无线、电线、光缆、RF等等，或者上述的任意合适的组合。Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including - but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
可以以一种或多种程序设计语言或其组合来编写用于执行本发明操作的计算机程序代码，所述程序设计语言包括面向对象的程序设计语言—诸如Java、Smalltalk、C++，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或终端上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络——包括局域网(LAN)或广域网(WAN)—连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。Computer program code for carrying out the operations of the present invention may be written in one or more programming languages, or combinations thereof, including object-oriented programming languages—such as Java, Smalltalk, C++, and conventional Procedural Programming Language—such as "C" or a similar programming language. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or terminal. In cases involving a remote computer, the remote computer may be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (such as through an Internet Service Provider). Internet connection).
注意，上述仅为本发明的较佳实施例及所运用技术原理。本领域技术人员会理解，本发明不限于这里所述的特定实施例，对本领域技术人员来说能够进行各种明显的变化、重新调整和替代而不会脱离本发明的保护范围。因此，虽然通过以上实施例对本发明进行了较为详细的说明，但是本发明不仅仅限于以上实施例，在不脱离本发明构思的情况下，还可以包括更多其他等效实施例，而本发明的范围由所附的权利要求范围决定。Note that the above are only preferred embodiments of the present invention and applied technical principles. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the protection scope of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and can also include more other equivalent embodiments without departing from the concept of the present invention, and the present invention The scope is determined by the scope of the appended claims.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910640133.8ACN110363590A (en) | 2019-07-16 | 2019-07-16 | Advertisement recommendation method, device, terminal and storage medium |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910640133.8ACN110363590A (en) | 2019-07-16 | 2019-07-16 | Advertisement recommendation method, device, terminal and storage medium |
| Publication Number | Publication Date |
|---|---|
| CN110363590Atrue CN110363590A (en) | 2019-10-22 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910640133.8APendingCN110363590A (en) | 2019-07-16 | 2019-07-16 | Advertisement recommendation method, device, terminal and storage medium |
| Country | Link |
|---|---|
| CN (1) | CN110363590A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110782286A (en)* | 2019-10-25 | 2020-02-11 | 上海喜马拉雅科技有限公司 | Advertising push method, device, server and computer-readable storage medium |
| CN111028005A (en)* | 2019-11-29 | 2020-04-17 | 腾讯科技（深圳）有限公司 | Advertisement display method and device |
| CN111176965A (en)* | 2019-12-19 | 2020-05-19 | 广州华多网络科技有限公司 | Pre-release test method and device for recommendation system and electronic equipment |
| CN111311328A (en)* | 2020-02-20 | 2020-06-19 | 支付宝(杭州)信息技术有限公司 | Method and device for determining advertisement click rate of product under advertisement channel |
| CN111833096A (en)* | 2020-06-10 | 2020-10-27 | 北京龙云科技有限公司 | Advertisement recommendation method and system based on machine learning |
| CN111861623A (en)* | 2019-12-30 | 2020-10-30 | 北京骑胜科技有限公司 | Information recommendation method, device and equipment |
| CN112036951A (en)* | 2020-09-03 | 2020-12-04 | 猪八戒股份有限公司 | Business opportunity recommendation method, system, electronic device and medium based on CNN model |
| CN112488759A (en)* | 2020-11-30 | 2021-03-12 | 上海酷量信息技术有限公司 | Advertisement sorting system and method |
| CN112785069A (en)* | 2021-01-29 | 2021-05-11 | 北京明略昭辉科技有限公司 | Prediction method and device for terminal equipment changing machine, storage medium and electronic equipment |
| CN113065896A (en)* | 2021-03-29 | 2021-07-02 | 上海酷量信息技术有限公司 | Advertisement recommendation method and device |
| CN113344648A (en)* | 2021-08-05 | 2021-09-03 | 北京龙云科技有限公司 | Advertisement recommendation method and system based on machine learning |
| CN113919856A (en)* | 2020-07-09 | 2022-01-11 | 上海钧正网络科技有限公司 | Target user selection method, system, device and storage medium |
| CN114140162A (en)* | 2021-12-03 | 2022-03-04 | 深圳乐信软件技术有限公司 | Advertisement delivery method, apparatus, electronic device, and computer-readable storage medium |
| WO2022089020A1 (en)* | 2020-10-30 | 2022-05-05 | 京东方科技集团股份有限公司 | Method and device for event displaying, storage medium, and electronic device |
| CN114862447A (en)* | 2022-04-08 | 2022-08-05 | 无线生活（杭州）信息科技有限公司 | Advertisement reordering method and device |
| CN114996575A (en)* | 2022-05-30 | 2022-09-02 | 北京达佳互联信息技术有限公司 | Resource information pushing method and device, electronic equipment and storage medium |
| CN115049445A (en)* | 2022-08-17 | 2022-09-13 | 铅笔头(深圳)科技有限公司 | Advertisement production evaluation system and method based on influence indexes |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100250335A1 (en)* | 2009-03-31 | 2010-09-30 | Yahoo! Inc | System and method using text features for click prediction of sponsored search advertisements |
| CN104965890A (en)* | 2015-06-17 | 2015-10-07 | 深圳市腾讯计算机系统有限公司 | Advertisement recommendation method and apparatus |
| CN107105031A (en)* | 2017-04-20 | 2017-08-29 | 北京京东尚科信息技术有限公司 | Information-pushing method and device |
| CN108416649A (en)* | 2018-02-05 | 2018-08-17 | 北京三快在线科技有限公司 | Search result ordering method, device, electronic equipment and storage medium |
| CN109558544A (en)* | 2018-12-12 | 2019-04-02 | 拉扎斯网络科技（上海）有限公司 | Sorting method and device, server and storage medium |
| CN109615411A (en)* | 2018-10-29 | 2019-04-12 | 中国平安人寿保险股份有限公司 | Algorithmic model-based advertising delivery method and device, and electronic equipment |
| CN110008399A (en)* | 2019-01-30 | 2019-07-12 | 阿里巴巴集团控股有限公司 | A kind of training method and device, a kind of recommended method and device of recommended models |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100250335A1 (en)* | 2009-03-31 | 2010-09-30 | Yahoo! Inc | System and method using text features for click prediction of sponsored search advertisements |
| CN104965890A (en)* | 2015-06-17 | 2015-10-07 | 深圳市腾讯计算机系统有限公司 | Advertisement recommendation method and apparatus |
| CN107105031A (en)* | 2017-04-20 | 2017-08-29 | 北京京东尚科信息技术有限公司 | Information-pushing method and device |
| CN108416649A (en)* | 2018-02-05 | 2018-08-17 | 北京三快在线科技有限公司 | Search result ordering method, device, electronic equipment and storage medium |
| CN109615411A (en)* | 2018-10-29 | 2019-04-12 | 中国平安人寿保险股份有限公司 | Algorithmic model-based advertising delivery method and device, and electronic equipment |
| CN109558544A (en)* | 2018-12-12 | 2019-04-02 | 拉扎斯网络科技（上海）有限公司 | Sorting method and device, server and storage medium |
| CN110008399A (en)* | 2019-01-30 | 2019-07-12 | 阿里巴巴集团控股有限公司 | A kind of training method and device, a kind of recommended method and device of recommended models |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110782286B (en)* | 2019-10-25 | 2023-08-18 | 上海喜马拉雅科技有限公司 | Advertisement pushing method, advertisement pushing device, server and computer readable storage medium |
| CN110782286A (en)* | 2019-10-25 | 2020-02-11 | 上海喜马拉雅科技有限公司 | Advertising push method, device, server and computer-readable storage medium |
| CN111028005B (en)* | 2019-11-29 | 2021-11-05 | 腾讯科技（深圳）有限公司 | Advertisement display method and device |
| CN111028005A (en)* | 2019-11-29 | 2020-04-17 | 腾讯科技（深圳）有限公司 | Advertisement display method and device |
| CN111176965A (en)* | 2019-12-19 | 2020-05-19 | 广州华多网络科技有限公司 | Pre-release test method and device for recommendation system and electronic equipment |
| CN111176965B (en)* | 2019-12-19 | 2023-10-13 | 广州方硅信息技术有限公司 | Recommendation system pre-release test method and device and electronic equipment |
| CN111861623A (en)* | 2019-12-30 | 2020-10-30 | 北京骑胜科技有限公司 | Information recommendation method, device and equipment |
| CN111311328B (en)* | 2020-02-20 | 2022-07-01 | 支付宝(杭州)信息技术有限公司 | Method and device for determining advertisement click rate of product under advertisement channel |
| CN111311328A (en)* | 2020-02-20 | 2020-06-19 | 支付宝(杭州)信息技术有限公司 | Method and device for determining advertisement click rate of product under advertisement channel |
| CN111833096A (en)* | 2020-06-10 | 2020-10-27 | 北京龙云科技有限公司 | Advertisement recommendation method and system based on machine learning |
| CN113919856A (en)* | 2020-07-09 | 2022-01-11 | 上海钧正网络科技有限公司 | Target user selection method, system, device and storage medium |
| CN112036951A (en)* | 2020-09-03 | 2020-12-04 | 猪八戒股份有限公司 | Business opportunity recommendation method, system, electronic device and medium based on CNN model |
| WO2022089020A1 (en)* | 2020-10-30 | 2022-05-05 | 京东方科技集团股份有限公司 | Method and device for event displaying, storage medium, and electronic device |
| US12437543B2 (en) | 2020-10-30 | 2025-10-07 | Boe Technology Group Co., Ltd. | Method and device for event displaying, storage medium, and electronic device |
| CN112488759A (en)* | 2020-11-30 | 2021-03-12 | 上海酷量信息技术有限公司 | Advertisement sorting system and method |
| CN112785069A (en)* | 2021-01-29 | 2021-05-11 | 北京明略昭辉科技有限公司 | Prediction method and device for terminal equipment changing machine, storage medium and electronic equipment |
| CN113065896A (en)* | 2021-03-29 | 2021-07-02 | 上海酷量信息技术有限公司 | Advertisement recommendation method and device |
| CN113344648A (en)* | 2021-08-05 | 2021-09-03 | 北京龙云科技有限公司 | Advertisement recommendation method and system based on machine learning |
| CN114140162A (en)* | 2021-12-03 | 2022-03-04 | 深圳乐信软件技术有限公司 | Advertisement delivery method, apparatus, electronic device, and computer-readable storage medium |
| CN114862447A (en)* | 2022-04-08 | 2022-08-05 | 无线生活（杭州）信息科技有限公司 | Advertisement reordering method and device |
| CN114996575A (en)* | 2022-05-30 | 2022-09-02 | 北京达佳互联信息技术有限公司 | Resource information pushing method and device, electronic equipment and storage medium |
| CN115049445A (en)* | 2022-08-17 | 2022-09-13 | 铅笔头(深圳)科技有限公司 | Advertisement production evaluation system and method based on influence indexes |
| Publication | Publication Date | Title |
|---|---|---|
| CN110363590A (en) | Advertisement recommendation method, device, terminal and storage medium | |
| KR102521785B1 (en) | Electronic commerce method for overseas purchasing and apparatus therefor | |
| CN110415032A (en) | Method, device, equipment and storage medium for estimating exposure conversion rate | |
| CN108256098B (en) | Method and device for determining emotional tendency of user comment | |
| US20240202800A1 (en) | Method, apparatus, device, storage medium and program product for object determination | |
| US20130080928A1 (en) | Embeddable context sensitive chat system | |
| CN106982256A (en) | Information-pushing method, device, equipment and storage medium | |
| CN109213936B (en) | Commodity searching method and device | |
| WO2017190610A1 (en) | Target user orientation method and device, and computer storage medium | |
| CN102541971A (en) | Mapping advertiser intents to keywords | |
| CN108777701A (en) | A kind of method and device of determining receiver | |
| CN108876545A (en) | Order recognition methods, device and readable storage medium storing program for executing | |
| WO2022151923A1 (en) | Method and apparatus for processing paperwork for goods, electronic device, medium, and program | |
| KR20210153817A (en) | System for mediating individual data of user | |
| CN112015970A (en) | Product recommendation method, related equipment and computer storage medium | |
| US10643276B1 (en) | Systems and computer-implemented processes for model-based underwriting | |
| CN113935780A (en) | Client loss risk prediction method based on survival analysis and related equipment thereof | |
| JP2022180274A (en) | Information processing system | |
| CN110942392A (en) | Service data processing method, device, equipment and medium | |
| CN112287208A (en) | User portrait generation method and device, electronic equipment and storage medium | |
| CN111754287B (en) | Article screening method, apparatus, device and storage medium | |
| CN114581140A (en) | Recommendation method and device for advertising floor and electronic equipment | |
| US20190180294A1 (en) | Supplier consolidation based on acquisition metrics | |
| CN105938606B (en) | Apparatus and method for providing account book service | |
| CN108460049A (en) | A kind of method and system of determining information category |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | Application publication date:20191022 | |
| RJ01 | Rejection of invention patent application after publication |