Advertisement information pushing method and device and electronic equipment
By obtaining the tag information of the target user and determining the advertising order based on the recall relationship, the problem of low accuracy of advertising content in the prior art is solved, and a more efficient advertising recall effect is achieved.
Patent Information
- Application Number
- CN202411598153.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-11
- Publication Date
- 2025-02-28
AI Technical Summary
In the prior art, when tag recall, the accuracy of advertising content is low, resulting in poor label recall effect.
By obtaining the tag information of the target user, based on the recall relationship between the user tag and the advertising order, the advertising order of the tag recall of the target user is determined, and these advertising information are pushed to the target user. The method includes training the target sorting model to sort advertising orders, sorting according to the expected revenue size, and pushing the advertising information with the largest expected revenue.
It improves the accuracy of advertising push, enhances the effect of advertising recall, and makes the advertising content more accurately match user needs.
Smart Images
Figure CN119539882A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a method, device and electronic device for pushing advertising information. Background Art
[0002] Advertisements and recommendation systems are core components of modern Internet services. They improve user experience and business benefits by showing personalized content to users. Tag Recall is the process of recalling ad content that matches user tags from the ad library based on user tags during ad delivery. This allows ads related to users to be quickly screened out from a large number of ad candidate sets for display.
[0003] Currently, during label recall, advertisements are pushed based on the similarity between user labels and advertisement content. The accuracy of advertisement content pushed in this way is low, resulting in poor label recall effect. Summary of the invention
[0004] The embodiments of the present application provide a method, device and electronic device for pushing advertising information to solve the problem that the accuracy of pushed advertising content is low, resulting in poor label recall effect.
[0005] To solve the above technical problems, this application is implemented as follows:
[0006] In a first aspect, an embodiment of the present application provides a method for pushing advertising information, the method comprising:
[0007] When receiving an access request from a target user for a target interface, obtaining information of a currently logged-in target account, and determining M tags of the target user based on the information of the target account;
[0008] Determine N advertising orders recalled by the M tags according to the recall relationship between the user tags and the advertising orders obtained in advance, wherein the recall relationship is determined according to advertising orders historically recalled by the user tags, and both M and N are positive integers;
[0009] The advertisement information of the N advertisement orders is pushed to the target account, where the advertisement information is information for indicating the advertisement contents corresponding to the N advertisement orders, and the advertisement information is for displaying on the target interface.
[0010] Optionally, before pushing the advertisement information corresponding to the N advertisement orders to the target account, the method further includes:
[0011] Inputting the advertisement order data in the access request log into a target ranking model for training, wherein the advertisement order data is data used to represent the characteristics of the advertisement order, and the target ranking model is used to sort the advertisement orders corresponding to the advertisement order data according to the expected revenue size;
[0012] Inputting the advertisement order data of the advertisement orders corresponding to the user tags in the user tag set into the trained target ranking model to obtain a score for each advertisement order, wherein the score is used to represent the expected revenue of each advertisement order;
[0013] sorting each of the advertising orders according to the size of the scores;
[0014] The pushing the advertisement information corresponding to the N advertisement orders to the target account includes:
[0015] At least one advertisement order that is ranked first is obtained from the N advertisement orders, and advertisement information corresponding to the at least one advertisement order is pushed to the target account.
[0016] Optionally, before determining the N advertising orders recalled by the M tags according to the recall relationship between the pre-acquired user tags and the advertising orders, the method further includes:
[0017] Determine the advertisement position information of the target interface, wherein the advertisement position information is position information in the target interface for displaying advertisements;
[0018] At the location indicated by the advertisement location information, obtaining multiple advertisement orders corresponding to multiple advertisements displayed historically;
[0019] The determining, based on the recall relationship between the pre-acquired user tags and the advertising orders, the N advertising orders recalled by the M tags includes:
[0020] Among the multiple advertising orders, N advertising orders recalled by the M tags are determined according to the recall relationship between the pre-acquired user tags and the advertising orders.
[0021] Optionally, before determining the N advertising orders recalled by the M tags according to the recall relationship between the pre-acquired user tags and the advertising orders, the method further includes:
[0022] Calculate the expected revenue brought by the advertising order corresponding to each user tag according to the advertising behavior data of the user corresponding to the user tag, wherein the advertising behavior data is the behavior data generated by the user corresponding to each user tag in the process of interacting with the advertisement, and the expected revenue is obtained by converting the advertising behavior data into revenue according to a preset conversion method;
[0023] Obtain multiple user tags corresponding to multiple expected benefits with the largest benefits from the expected benefits to obtain a user tag set;
[0024] Obtaining a historically recalled advertising order for each user tag in the user tag set;
[0025] According to the advertisement orders historically recalled by each user tag, a recall relationship between each user tag and the advertisement order is determined.
[0026] Optionally, calculating the expected revenue brought by the advertising order corresponding to each user tag according to the advertising behavior data of the user corresponding to the user tag includes:
[0027] Acquire advertising behavior data associated with each user tag and a target combination attribute of the advertisement, wherein the target combination attribute is one of a plurality of combination attributes obtained by combining at least one position attribute and at least one content attribute of the advertisement, wherein the position attribute is an attribute associated with the position of the advertisement, and the content attribute is an attribute associated with the content of the advertisement;
[0028] Calculating, according to the advertising behavior data corresponding to the target combination attribute, the expected revenue brought by the advertising order corresponding to each tag based on the target combination attribute;
[0029] The user tag set is obtained by obtaining a plurality of user tags corresponding to a plurality of expected benefits with the largest benefits among the expected benefits, and includes:
[0030] Among the expected benefits corresponding to each combination attribute of the advertising order, multiple user tags corresponding to multiple expected benefits with the largest benefits are obtained to obtain the user tag set.
[0031] Optionally, the user tag set is obtained by obtaining multiple user tags corresponding to multiple expected benefits with the largest benefits among the expected benefits, including:
[0032] According to the relative size of the expected revenue brought by the advertisement orders corresponding to each user tag and the average revenue, determining the ratio of the expected revenue brought by the advertisement orders corresponding to each user tag to the average revenue, wherein the average revenue is the average revenue value brought by the advertisement orders within the target range;
[0033] Sort the user tags corresponding to the expected benefits in descending order of the proportions;
[0034] A plurality of user tags corresponding to the plurality of expected benefits that are ranked first are obtained to obtain the user tag set.
[0035] In a second aspect, an embodiment of the present application provides a device for pushing advertising information, the device comprising:
[0036] A first acquisition module is used to, upon receiving an access request from a target user to a target interface, acquire information of a currently logged-in target account, and determine M tags of the target user based on the information of the target account;
[0037] A first determination module is used to determine the N advertising orders recalled by the M tags according to the recall relationship between the user tags and the advertising orders obtained in advance, wherein the recall relationship is determined according to the advertising orders historically recalled by the user tags, and both M and N are positive integers;
[0038] A push module is used to push the advertisement information of the N advertisement orders to the target account, wherein the advertisement information is information used to represent the advertisement content corresponding to the N advertisement orders, and the advertisement information is used to be displayed on the target interface.
[0039] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method for pushing advertising information described in the first aspect.
[0040] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for pushing advertising information described in the first aspect are implemented.
[0041] In a fifth aspect, a computer program product is provided, comprising computer instructions, which, when executed by a processor, implement the steps of the method for pushing advertising information as described in the first aspect.
[0042] In an embodiment of the present application, an advertising order is determined based on the recall relationship between the user tag of the target user and the advertising order, and is pushed to the target user. This can more accurately match the user's needs and advertising content, improve the effect of obtaining the user's advertising preferences, and improve the accuracy of advertising order push, thereby improving the accuracy and effect of advertising recall. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0044] Figure 1This is one of the flow charts of a method for pushing advertising information provided in an embodiment of the present application;
[0045] Figure 2 This is the second flowchart of a method for pushing advertising information provided in an embodiment of the present application;
[0046] Figure 3 It is a structural diagram of a device for pushing advertising information provided by an embodiment of the present application;
[0047] Figure 4 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0048] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0049] The embodiments of the present application provide a method, device and electronic device for pushing advertising information to solve the problem that the accuracy of pushed advertising content is low, resulting in poor label recall effect.
[0050] See also Figure 1 , Figure 1 is a flow chart of a method for pushing advertising information provided by an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:
[0051] Step 101: upon receiving an access request from a target user for a target interface, obtaining information of a currently logged-in target account, and determining M tags of the target user based on the information of the target account;
[0052] Step 102: Determine N advertising orders recalled by the M tags based on the recall relationship between the user tags and the advertising orders obtained in advance, wherein the recall relationship is determined based on advertising orders historically recalled by the user tags, and both M and N are positive integers;
[0053] Step 103: Push the advertisement information of the N advertisement orders to the target account, where the advertisement information is information for indicating the advertisement contents corresponding to the N advertisement orders, and the advertisement information is for displaying on the target interface.
[0054] The target account may be a virtual account corresponding to the target user on the application. When the target account is logged in, when the target user operates on the target interface, the target user's operation behavior data may be obtained, including browsing records, access records, etc., and the target user's attribute information and other information may also be obtained, thereby forming the target account information.
[0055] The label of the target user can be determined based on the information of the target account. The label of the target user may include one or more labels, which may include the attribute label of the target user, the advertising behavior label, the model prediction class label, etc.
[0056] Among them, the attribute labels of the target users may be labels used to reflect the user attributes, and these labels are mainly constructed based on the demographic portrait data of the target users, for example, the age group, geographical range, and life stage of the target users.
[0057] The target user's advertising behavior label can be understood as the target user's advertising-related behaviors, including the target user's exposure, clicks, conversions, and payments on advertising, industry, media, and other related content. For example, the target user clicks, pays, and downloads game ads while browsing them.
[0058] The model prediction class label can be a prediction label made based on the target user's behavior data in the industry, or a user label determined by the looklike technology. For example, the characteristics of a specific group of people are obtained, and based on the characteristics of the specific group of people, potential user groups with similar characteristics are identified and expanded, thereby obtaining the label of the potential user group.
[0059] Through the above method, one or more tags of the target user can be obtained.
[0060] Before obtaining the advertising orders recalled by the target user's tag, the recall relationship between the user tag and the advertising order can be determined based on the advertising orders historically recalled by the user tag. When an advertisement is pushed to the user corresponding to the user tag based on the characteristics of the user tag, the advertising order corresponding to the advertisement is the advertising order recalled by the user tag.
[0061] For example, user group A often browses a specific type of product, such as "mountaineering equipment". The tag of this user group includes "outdoor enthusiast". The advertiser pushes advertisements for mountaineering equipment to the account corresponding to the tag. Based on the above process, a recall relationship between outdoor enthusiasts (user tags) and advertising orders for mountaineering equipment (advertising orders) can be established. The advertising orders for mountaineering equipment can include one or more orders.
[0062] When there is one tag, one tag can recall one or more advertising orders. When there are multiple tags, multiple tags can recall one or more advertising orders.
[0063] The advertising information of N advertising orders is pushed to the target account. In the specific implementation, part or all of the N advertising orders can be obtained as the target advertising orders according to the actual situation, and the advertising information of the target advertising orders can be pushed.
[0064] When obtaining some of the N advertising orders, one implementation method is to calculate at least one advertising order with the largest expected revenue according to the expected revenue of each advertising order, and push the advertising information of the at least one advertising order. For example, a sorting model is used to sort the advertising orders according to the expected revenue, so as to obtain one or more advertising orders ranked in the top few positions, and push the advertising information corresponding to the target advertising order to the target account. In other implementations, several advertising orders with the largest historical revenue can also be used as target advertising orders; or according to the number of advertising positions, several advertising orders with the largest historical exposure, clicks, and purchases can be determined as target advertising orders.
[0065] The advertisement information of N advertisement orders is displayed on the target interface, so that the user can quickly obtain the contents of the N advertisement orders through the advertisement information.
[0066] In the embodiment of the present application, based on the recall relationship between user tags and advertising orders, the advertising orders recalled by the user tags of the target users are determined, and the advertising information of the advertising orders is pushed to the target users, which can improve the accuracy of advertising push and thus improve the advertising recall effect.
[0067] Optionally, before pushing the advertisement information corresponding to the N advertisement orders to the target account, the method further includes:
[0068] Inputting the advertisement order data in the access request log into a target ranking model for training, wherein the advertisement order data is data used to represent the characteristics of the advertisement order, and the target ranking model is used to sort the advertisement orders corresponding to the advertisement order data according to the expected revenue size;
[0069] Inputting the advertisement order data of the advertisement orders corresponding to the user tags in the user tag set into the trained target ranking model to obtain a score for each advertisement order, wherein the score is used to represent the expected revenue of each advertisement order;
[0070] sorting each of the advertising orders according to the score;
[0071] The pushing the advertisement information corresponding to the N advertisement orders to the target account includes:
[0072] At least one advertisement order that is ranked first is obtained from the N advertisement orders, and advertisement information corresponding to the at least one advertisement order is pushed to the target account.
[0073] In this embodiment, the advertisement order data in the request log may be obtained, and the advertisement order data includes features of the advertisement order, such as advertisement attribute features, user behavior features, and other features of the advertisement order. The advertisement order data is input into the target ranking model for training, and the trained target ranking model can score the input advertisement orders and sort them according to the expected revenue size.
[0074] For example, the advertising order data in the obtained request log is used to construct positive and negative sample pairs according to whether it enters the fine ranking. A RankNet model is constructed, and a multi-layer fully connected network is constructed using the data on the order side. The pairwise loss function is used in the training process to maximize the distance between positive and negative samples. After the model training is completed, the model is used to predict the advertising orders in the advertising order set corresponding to the label set, and the ranking score is output, which can be used to represent the expected revenue size of the advertising order.
[0075] After the advertising orders are sorted from large to small according to the expected revenue values, multiple advertising orders that are sorted first, that is, topP advertising orders, are obtained, where P can be configured according to the specific situation and P is an integer greater than or equal to 1, so that the advertising information of the P advertising orders with the largest expected revenues is obtained for push.
[0076] In the above way, the order sorting is learned based on the sorting log of access requests, which reflects the personalization of labels and avoids the recall of popular orders caused by using only exposure data. The trained target sorting model is used to sort and score each advertising order in the advertising order set according to the expected revenue size, thereby obtaining the expected revenue size sorting of the advertising orders.
[0077] By obtaining the top advertising orders with larger expected advertising order revenues in the above manner and pushing the top advertising orders to the target accounts, the advertising effect can be improved.
[0078] Optionally, before determining the N advertising orders recalled by the M tags according to the recall relationship between the pre-acquired user tags and the advertising orders, the method further includes:
[0079] Determine the advertisement position information of the target interface, wherein the advertisement position information is position information in the target interface for displaying advertisements;
[0080] At the location indicated by the advertisement location information, obtaining multiple advertisement orders corresponding to multiple advertisements displayed historically;
[0081] The determining, based on the recall relationship between the pre-acquired user tags and the advertising orders, the N advertising orders recalled by the M tags includes:
[0082] Among the multiple advertising orders, N advertising orders recalled by the M tags are determined according to the recall relationship between the pre-acquired user tags and the advertising orders.
[0083] The advertisement location information may include information about the display location of the advertisement on the target interface, which may include coordinate information of the advertisement display area, platform information where the advertisement display area is located, and other information used to indicate the advertisement display area. In addition, the advertisement location information may also include the advertisement display method corresponding to the advertisement display area.
[0084] The advertisement orders of the multiple advertisements displayed historically are obtained at the position indicated by the advertisement position information. For example, the advertisement orders of the multiple advertisements displayed in the top area of the homepage of website A are obtained.
[0085] When there are multiple, that is, at least two, advertisement orders corresponding to the advertisement position information, N advertisement orders recalled by M tags are determined from the multiple advertisement orders.
[0086] In this way, N advertising orders associated with the advertising position and M tags can be obtained, and the N advertising orders can be pushed to the target user. When the N advertising orders are displayed at the location where the advertising position information is located, the recall effect of the advertising orders can be improved.
[0087] Optionally, before determining the N advertising orders recalled by the M tags according to the recall relationship between the pre-acquired user tags and the advertising orders, the method further includes:
[0088] Calculate the expected revenue brought by the advertising order corresponding to each user tag according to the advertising behavior data of the user corresponding to the user tag, wherein the advertising behavior data is the behavior data generated by the user corresponding to each user tag in the process of interacting with the advertisement, and the expected revenue is obtained by converting the advertising behavior data into revenue according to a preset conversion method;
[0089] Obtain multiple user tags corresponding to multiple expected benefits with the largest benefits from the expected benefits to obtain a user tag set;
[0090] Obtaining a historically recalled advertising order for each user tag in the user tag set;
[0091] According to the advertisement orders historically recalled by each user tag, a recall relationship between each user tag and the advertisement order is determined.
[0092] In this embodiment, the behavioral data generated by the user corresponding to each tag when interacting with the advertisement is obtained, including the behavioral data corresponding to the exposure, click, conversion, payment and other behavioral operations of the account corresponding to the tag. The above-mentioned behavioral data includes the behavioral data corresponding to the tag in the traffic dimension and / or industry dimension. Based on the above-mentioned behavioral data, the expected revenue of the advertising order corresponding to each tag or each tag combination is calculated respectively. The expected revenue is the conversion of the advertising behavior data into revenue according to the preset conversion method. For example, according to the exposure, click, conversion, payment and other behavioral operations of the user corresponding to the tag, the cost per thousand impressions (Revenue Per Mille, RPM), deep expected RPM and return on investment (Return on Investment, ROI) of each tag in the traffic attribute cross-advertising attribute dimension are calculated. These data can reflect the expected revenue of the advertising order corresponding to each tag.
[0093] Based on the calculated expected revenue of the advertising orders corresponding to each tag, multiple tags corresponding to the advertising orders with the highest expected revenue are obtained in descending order of expected revenue as the user tag set, that is, multiple user tags corresponding to the advertising orders with the largest expected revenue.
[0094] Through the above method, it is possible to obtain high-quality user labels in different dimensions, for example, a set of high-quality user labels in the traffic cross-advertising dimension.
[0095] By obtaining a user tag set in the above manner, user tags can be stratified to obtain high-quality tags with better recall effects, which can improve the advertising recall effect. Tags with poor effects can be automatically eliminated, and new tags can be automatically associated without manual operation, which improves the effect of tag recall advertising orders. This solves the problem of low efficiency and poor effect caused by manual matching based on rules and experience in traditional tags.
[0096] According to the advertisement order data of the user tag history recall, the advertisement order of each tag history recall in the user tag set is obtained, wherein the advertisement order of the tag history recall can refer to the description corresponding to the part of establishing the recall relationship between the user tag and the advertisement order.
[0097] For the advertising order recalled by a tag, taking the target tag in the user tag set as an example, in one implementation, the advertising order recalled by the target tag is obtained, and the order is the advertising order recalled by the target tag.
[0098] In another implementation, when the target tag in the user tag set is included in the user tag set, the advertising orders recalled by the tag set are determined as the advertising orders recalled by the target tag. For example, the tag set corresponding to account A includes tags A, B, and C, and the advertising orders recalled by the tag set A, B, and C can be determined as the advertising orders recalled by tag A.
[0099] When determining the set of advertisement orders recalled by the user tag set, in one implementation, the advertisement order recalled by each tag is determined, and the advertisement order set is determined based on the advertisement order recalled by each tag.
[0100] In another implementation, at least two tags in the user tag set are combined to obtain multiple tag combinations, and the recalled advertising order set is determined based on each tag combination in the multiple tag combinations. For example, any two or more tags in the user tag set can be combined respectively, and the tag combination can include two, three or more tags. The advertising order set is determined based on the advertising orders recalled by the tag combination.
[0101] According to the recall relationship between advertising orders and user tags, the tags are automatically associated with the recall orders, which reduces manual evaluation and configuration operations and improves the configuration efficiency during the advertising recall process.
[0102] Optionally, calculating the expected revenue brought by the advertising order corresponding to each user tag according to the advertising behavior data of the user corresponding to the user tag includes:
[0103] Acquire advertising behavior data associated with each user tag and a target combination attribute of the advertisement, wherein the target combination attribute is one of a plurality of combination attributes obtained by combining at least one position attribute and at least one content attribute of the advertisement, wherein the position attribute is an attribute associated with the position of the advertisement, and the content attribute is an attribute associated with the content of the advertisement;
[0104] Calculating, according to the advertising behavior data corresponding to the target combination attribute, the expected revenue brought by the advertising order corresponding to each tag based on the target combination attribute;
[0105] The user tag set is obtained by obtaining a plurality of user tags corresponding to a plurality of expected benefits with the largest benefits among the expected benefits, and includes:
[0106] Among the expected benefits corresponding to each combination attribute of the advertising order, multiple user tags corresponding to multiple expected benefits with the largest benefits are obtained to obtain the user tag set.
[0107] When the number of attributes of the location attribute of the advertisement includes at least one, and the number of attributes of the content attribute of the advertisement also includes at least one, any location attribute and any advertisement content attribute can be obtained and combined to obtain a target attribute combination, and the target attribute combination can be one of the attribute combinations among multiple attribute combinations. Among them, the advertisement location attribute can be an attribute related to the advertisement location, including the platform where the advertisement is located, the advertisement position, the advertisement source (obtained by clicking on a specific website or link), etc. The advertisement content attributes can include attributes in dimensions such as the industry to which the advertisement belongs, the corresponding customer, account, and promotional product.
[0108] The advertisement behavior data associated with the user tag and the target combination attribute of the advertisement is obtained, so that the advertisement behavior data of the user tag at the intersection of the position attribute and the content attribute can be obtained.
[0109] Based on the above behavioral data, the expected revenue related to the advertising orders corresponding to each tag and the target combination attributes is calculated.
[0110] For example, a user browses an advertisement for mountaineering equipment on platform A and clicks on the advertisement. Based on the advertising behavior data corresponding to the click operation, the expected revenue generated by orders on platform A and the outdoor industry for users with the label "outdoor enthusiast" is calculated.
[0111] By combining each location attribute with each content attribute, multiple combined attributes can be obtained. The expected benefits of each user tag based on the target combined attribute are obtained, and multiple maximum expected benefits are obtained. Multiple user tags corresponding to the multiple maximum expected benefits form a user tag set. In this way, high-quality tags under the dimension of the target combined attribute, that is, a user tag set, can be obtained.
[0112] Through the above method, the user's advertising behavior data on the target combination attribute dimension can be obtained more accurately, so as to calculate the expected profit corresponding to the label based on the advertising behavior data, which can improve the accuracy of advertising behavior data acquisition and thus improve the effect of advertising recall.
[0113] Optionally, the user tag set is obtained by obtaining multiple user tags corresponding to multiple expected benefits with the largest benefits among the expected benefits, including:
[0114] According to the relative size of the expected revenue brought by the advertisement orders corresponding to each user tag and the average revenue, determining the ratio of the expected revenue brought by the advertisement orders corresponding to each user tag to the average revenue, wherein the average revenue is the average revenue value brought by the advertisement orders within the target range;
[0115] Sort the user tags corresponding to the expected benefits in descending order of the proportions;
[0116] A plurality of user tags corresponding to the plurality of expected benefits that are ranked first are obtained to obtain the user tag set.
[0117] When obtaining a user tag set, the expected revenue brought by the orders corresponding to each tag can be compared with the average revenue. The average revenue can be understood as the overall revenue, that is, the average revenue of advertising orders in the target dimension, such as the expected RPM, deep expected RPM value or ROI data in an industry.
[0118] The comparison of the expected revenue brought by the orders corresponding to each tag relative to the average revenue can be based on the same parameter comparison. For example, the first expected RPM of each tag in the cross-advertising attribute dimension of the traffic attribute dimension and the average expected RPM in the cross-advertising attribute dimension of the traffic attribute dimension is calculated, and the ratio of the first expected RPM to the average expected RPM is calculated.
[0119] According to the ratio of the expected revenue brought by the orders corresponding to each tag to the average revenue, the user tags corresponding to the ratio (or expected revenue) are sorted in order from large to small. The first multiple user tags are obtained to form a user tag set, that is, the topK, where K is configured according to the specific situation. The expected revenue brought by the orders corresponding to the user tags in the user tag set is higher, while the expected revenue brought by the orders corresponding to other user tags is lower, and they can be automatically eliminated. In this way, when the advertising effect of the tag deteriorates over time, it can be automatically eliminated without manual configuration, which can reduce manual operations and improve the tag recall effect.
[0120] In order to facilitate understanding of the above embodiments, examples are given below in combination with specific implementation methods.
[0121] The current label recall is mainly achieved through the customization of manual rules, which is a static rule implementation and offline effect evaluation. The problem with this implementation method is that when the label effect deteriorates over time, it cannot be updated in time to eliminate these labels. For new labels, manual configuration is required before they can take effect. In addition, labels are batched, and the amount of label rule configuration evaluation is huge. In the specific implementation process, manual configuration is more and the efficiency is low.
[0122] In order to solve the above problems, the automatic association between labels and orders is realized through label production module, label population stratification module, label recall module and order sorting module, and used for advertising recall strategy.
[0123] like Figure 2 As shown, the method for pushing advertising information includes the following steps:
[0124] Step 1: Generate user tags. This step is performed by the tag generation module.
[0125] User tags are constructed through mining, model prediction, and data purchase, including basic tags obtained from internal and external data, as well as tags for model prediction extension of these tags. Among them, user tags can be divided into user attribute tags, advertising behavior tags, and model prediction tags. User attribute tags and advertising behavior tags are generated by collecting user portraits and business statistical data. In addition, prediction tags and extended tags are generated according to the model, and the above tags are written into the user portrait table.
[0126] Step 2: Count the specific business effect data of different tags, and stratify the effect of tags in corresponding industries / accounts, that is, based on the improvement ratio of the industry / account of the tag in different traffic dimensions compared with the overall market in terms of business indicators, the level of tags is divided according to the improvement ratio data, and high-quality tags in each dimension are obtained. This step is performed through the tag population stratification module.
[0127] By obtaining the user's advertising behavior data and the user's tag information, the advertising behavior data can be obtained by counting the user's exposure, click, conversion, payment and other behavior data of the tag in the traffic dimension (traffic source, platform) and cross-advertising dimension (industry, customer, account, promoted product). Based on the above behavior data, the expected RPM, deep expected RPM and ROI data of the tag corresponding to the traffic cross-advertising dimension can be calculated.
[0128] At the same time, we can obtain the expected RPM, deep expected RPM, and ROI of the market (average), and calculate the improvement ratio of the above data of the label relative to the market data. Thus, we can get the ranking and stratification of the population labels in the traffic cross-advertising dimension, and finally output the improvement ratio of the label in each dimension relative to the market and the corresponding label. We can perform label stratification based on the improvement ratio and label information, and update the label population stratification table. Based on the population stratification table, we can get high-quality labels in each dimension.
[0129] Step 3: Read the high-quality labels in the population stratification table, obtain the high-quality labels of each industry based on the population label stratification results, and automatically associate the advertising order recall. Associate the labels and industries to automatically match the recall relationship between labels and orders in different industries. This step is performed by the label recall module.
[0130] First, according to the label population stratification table, sort from large to small according to the improvement ratio to obtain the topK with the largest improvement ratio. K can be configured according to the specific situation to obtain a set of high-quality labels in the traffic cross-advertising dimension.
[0131] Secondly, the data of the advertisement dimension are uniformly mapped (or refined) to specific advertisement orders, thereby obtaining a candidate set of advertisement orders associated with the tags.
[0132] Finally, summarize all candidate order sets associated with high-quality labels, deduplicate the orders in the candidate order set, and write them into the label order recall table.
[0133] Step 4: For the recalled orders associated with the tag, the corresponding sorting model is trained according to the request log to sort all the overall orders and return the top orders to complete the recall. This step is performed by the order sorting module.
[0134] All the orders recalled by labels are sorted by expected revenue, where the expected revenue can include the revenue already obtained in the past and the expected revenue that may be brought in the future based on the current situation. The scoring and sorting of all orders are completed by training the label order sorting model.
[0135] The construction logic of the label order sorting model is as follows: first, obtain the advertising order data in the request log, build positive and negative sample pairs according to whether to enter the fine ranking, and then build a RankNet model. Specifically, use the information on the order side to build a multi-layer fully connected network. The training process uses the pairwise loss function to maximize the distance between positive and negative samples. Finally, make predictions for the orders output by the label recall module, output the ranking score, and finally write it to the database (DB), and update the order score to the label order recall table for online access requests.
[0136] Through the above process, the entire system is completely decoupled from manual configuration and evaluation, achieving adaptive recall of high-quality labels.
[0137] Step 5: When a user makes an online access request, the corresponding advertising order is obtained from the tag sorting table according to the tag in the user's tag set, and the head orders are obtained in descending order according to the advertising order sorting score to fill the recall set until the recall quota is reached and the recall process ends.
[0138] When a user accesses an online site, he or she can obtain traffic information of the interface corresponding to the access request, such as the ad space on the current interface and the user's tag set.
[0139] The embodiments of the present application can achieve the following technical effects:
[0140] 1. Label population stratification can effectively establish the mapping relationship between labels in the industry, replacing the previous cumbersome process of manual configuration and manual evaluation, and simplifying the process of label access in recall configuration.
[0141] 2. The adaptive recall of labels can automatically eliminate low-quality labels and introduce new high-quality labels, effectively improving the recall effect of the entire label recall system.
[0142] 3. The entire system completely decouples label mining and label application, greatly improving the efficiency of label application.
[0143] See also Figure 3 , Figure 3 is a schematic diagram of a structure of a device for pushing advertising information provided by an embodiment of the present application, such as Figure 3 As shown, the advertising information pushing device 300 includes:
[0144] The first acquisition module 301 is used to acquire information of the currently logged-in target account when receiving an access request for a target interface from a target user, and determine M tags of the target user based on the information of the target account;
[0145] A first determination module 302 is used to determine N advertising orders recalled by the M tags according to the recall relationship between the user tags and the advertising orders obtained in advance, wherein the recall relationship is determined according to the advertising orders historically recalled by the user tags, and both M and N are positive integers;
[0146] The push module 303 is used to push the advertisement information of the N advertisement orders to the target account, wherein the advertisement information is information used to represent the advertisement content corresponding to the N advertisement orders, and the advertisement information is used to be displayed on the target interface.
[0147] Optionally, the device further comprises:
[0148] a training module, configured to input the advertisement order data in the access request log into a target ranking model for training, wherein the advertisement order data is data used to represent characteristics of the advertisement order, and the target ranking model is used to sort the advertisement orders corresponding to the advertisement order data according to the size of expected revenue;
[0149] An input module, used for inputting the advertisement order data of the advertisement order corresponding to the user tag in the user tag set into the trained target ranking model to obtain a score of each advertisement order, wherein the score is used to represent the expected revenue of each advertisement order;
[0150] A sorting module, used for sorting each of the advertising orders according to the size of the scores;
[0151] The push module is specifically used for:
[0152] At least one advertisement order that is ranked first is obtained from the N advertisement orders, and advertisement information corresponding to the at least one advertisement order is pushed to the target account.
[0153] Optionally, the device further comprises:
[0154] A second determination module is used to determine the advertisement position information of the target interface, wherein the advertisement position information is the position information for displaying advertisements in the target interface;
[0155] A second acquisition module is used to acquire, at the location indicated by the advertisement location information, multiple advertisement orders corresponding to multiple advertisements displayed historically;
[0156] The first determining module is specifically used for:
[0157] Among the multiple advertising orders, N advertising orders recalled by the M tags are determined according to the recall relationship between the pre-acquired user tags and the advertising orders.
[0158] Optionally, the device further comprises:
[0159] A calculation module, configured to calculate the expected revenue brought by the advertising order corresponding to each user tag according to the advertising behavior data of the user corresponding to the user tag, wherein the advertising behavior data is the behavior data generated by the user corresponding to each user tag in the process of interacting with the advertisement, and the expected revenue is obtained by converting the advertising behavior data into revenue according to a preset conversion method;
[0160] A third acquisition module is used to acquire multiple user tags corresponding to multiple expected benefits with the largest benefits from the expected benefits, and obtain a user tag set;
[0161] A fourth acquisition module is used to acquire the historically recalled advertising orders for each user tag in the user tag set;
[0162] The third determining module is used to determine the recall relationship between each user tag and the advertising order according to the advertising order historically recalled by each user tag.
[0163] Optionally, the calculation module includes:
[0164] A first acquisition submodule is used to acquire advertising behavior data associated with each user tag and a target combination attribute of the advertisement, wherein the target combination attribute is one of a plurality of combination attributes obtained by combining at least one position attribute and at least one content attribute of the advertisement, wherein the position attribute is an attribute associated with the position of the advertisement, and the content attribute is an attribute associated with the content of the advertisement;
[0165] A calculation submodule, configured to calculate, according to the advertising behavior data corresponding to the target combination attribute, the expected revenue brought by the advertising order corresponding to each tag based on the target combination attribute;
[0166] The third acquisition module is specifically used for:
[0167] Among the expected benefits corresponding to each combination attribute of the advertising order, multiple user tags corresponding to multiple expected benefits with the largest benefits are obtained to obtain the user tag set.
[0168] Optionally, the third acquisition module includes:
[0169] a determination submodule, configured to determine a ratio of the expected revenue brought by the advertising orders corresponding to each user tag to the average revenue according to the relative size of the expected revenue brought by the advertising orders corresponding to each user tag and the average revenue, wherein the average revenue is an average revenue value brought by the advertising orders within a target range;
[0170] A sorting submodule, used to sort the user tags corresponding to the expected benefits in descending order of the proportions;
[0171] The second acquisition submodule is used to acquire multiple user tags corresponding to the first multiple expected benefits, and obtain the user tag set.
[0172] The push device for advertising information can achieve Figure 1 The various processes implemented in the method embodiment can achieve the same technical effect, and to avoid repetition, they will not be described again here.
[0173] like Figure 4 As shown, an embodiment of the present application further provides an electronic device 400, comprising: a processor 401, a memory 402, and a program stored in the memory 402 and executable on the processor 401. When the program is executed by the processor 401, each process of the above-mentioned advertising information push method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.
[0174] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, each process of the above-mentioned method for pushing advertising information is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0175] The present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of the method embodiment shown can achieve the same technical effect, and will not be described again here to avoid repetition.
[0176] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.
[0177] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0178] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.
Claims
1. A method for pushing advertising information, characterized in that: include: When receiving an access request from a target user for a target interface, obtaining information of a currently logged-in target account, and determining M tags of the target user based on the information of the target account; Determine N advertising orders recalled by the M tags according to the recall relationship between the user tags and the advertising orders obtained in advance, wherein the recall relationship is determined according to advertising orders historically recalled by the user tags, and both M and N are positive integers; The advertisement information of the N advertisement orders is pushed to the target account, where the advertisement information is information for indicating the advertisement contents corresponding to the N advertisement orders, and the advertisement information is for displaying on the target interface.
2. The method according to claim 1, characterized in that: Before pushing the advertisement information corresponding to the N advertisement orders to the target account, the method further includes: Inputting the advertisement order data in the access request log into a target ranking model for training, wherein the advertisement order data is data used to represent the characteristics of the advertisement order, and the target ranking model is used to sort the advertisement orders corresponding to the advertisement order data according to the expected revenue size; Inputting the advertisement order data of the advertisement orders corresponding to the user tags in the user tag set into the trained target ranking model to obtain a score for each advertisement order, wherein the score is used to represent the expected revenue of each advertisement order; sorting each of the advertising orders according to the size of the scores; The pushing the advertisement information corresponding to the N advertisement orders to the target account includes: At least one advertisement order that is ranked first is obtained from the N advertisement orders, and advertisement information corresponding to the at least one advertisement order is pushed to the target account.
3. The method according to claim 1, characterized in that Before determining the N advertising orders recalled by the M tags according to the recall relationship between the pre-acquired user tags and the advertising orders, the method further includes: Determine the advertisement position information of the target interface, wherein the advertisement position information is position information in the target interface for displaying advertisements; At the location indicated by the advertisement location information, obtaining multiple advertisement orders corresponding to multiple advertisements displayed historically; The determining, based on the recall relationship between the pre-acquired user tags and the advertising orders, the N advertising orders recalled by the M tags includes: Among the multiple advertising orders, N advertising orders recalled by the M tags are determined according to the recall relationship between the pre-acquired user tags and the advertising orders.
4. The method according to any one of claims 1 to 3, characterized in that: Before determining the N advertising orders recalled by the M tags according to the recall relationship between the pre-acquired user tags and the advertising orders, the method further includes: Calculate the expected revenue brought by the advertising order corresponding to each user tag according to the advertising behavior data of the user corresponding to the user tag, wherein the advertising behavior data is the behavior data generated by the user corresponding to each user tag in the process of interacting with the advertisement, and the expected revenue is obtained by converting the advertising behavior data into revenue according to a preset conversion method; Obtain multiple user tags corresponding to multiple expected benefits with the largest benefits from the expected benefits to obtain a user tag set; Obtaining a historically recalled advertising order for each user tag in the user tag set; According to the advertisement orders historically recalled by each user tag, a recall relationship between each user tag and the advertisement order is determined.
5. The method according to claim 4, characterized in that The calculating, based on the advertising behavior data of the user corresponding to the user tag, the expected revenue brought by the advertising order corresponding to each user tag includes: Acquire advertising behavior data associated with each user tag and a target combination attribute of the advertisement, wherein the target combination attribute is one of a plurality of combination attributes obtained by combining at least one position attribute and at least one content attribute of the advertisement, wherein the position attribute is an attribute associated with the position of the advertisement, and the content attribute is an attribute associated with the content of the advertisement; Calculating, according to the advertising behavior data corresponding to the target combination attribute, the expected revenue brought by the advertising order corresponding to each tag based on the target combination attribute; The user tag set is obtained by obtaining a plurality of user tags corresponding to a plurality of expected benefits with the largest benefits among the expected benefits, and includes: Among the expected benefits corresponding to each combination attribute of the advertising order, multiple user tags corresponding to multiple expected benefits with the largest benefits are obtained to obtain the user tag set.
6. The method according to claim 4, characterized in that The user tag set is obtained by obtaining a plurality of user tags corresponding to a plurality of expected benefits with the largest benefits among the expected benefits, and includes: According to the relative size of the expected revenue brought by the advertisement orders corresponding to each user tag and the average revenue, determining the ratio of the expected revenue brought by the advertisement orders corresponding to each user tag to the average revenue, wherein the average revenue is the average revenue value brought by the advertisement orders within the target range; Sort the user tags corresponding to the expected benefits in descending order of the proportions; A plurality of user tags corresponding to the plurality of expected benefits that are ranked first are obtained to obtain the user tag set.
7. A device for pushing advertising information, characterized in that: include: A first acquisition module is used to, upon receiving an access request from a target user to a target interface, acquire information of a currently logged-in target account, and determine M tags of the target user based on the information of the target account; A first determination module is used to determine the N advertising orders recalled by the M tags according to the recall relationship between the user tags and the advertising orders obtained in advance, wherein the recall relationship is determined according to the advertising orders historically recalled by the user tags, and both M and N are positive integers; A push module is used to push the advertisement information of the N advertisement orders to the target account, wherein the advertisement information is information used to represent the advertisement content corresponding to the N advertisement orders, and the advertisement information is used to be displayed on the target interface.
8. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps of the method for pushing advertising information according to any one of claims 1 to 6 are implemented.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for pushing advertisement information according to any one of claims 1 to 6 are implemented.
10. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the method for pushing advertising information as claimed in any one of claims 1 to 6.