CN116342214A

CN116342214A - Service pushing and calibration parameter obtaining method and device

Info

Publication number: CN116342214A
Application number: CN202310120904.7A
Authority: CN
Inventors: 林翔; 王海东; 史峰
Original assignee: Rajax Network Technology Co Ltd
Current assignee: Rajax Network Technology Co Ltd
Priority date: 2023-02-14
Filing date: 2023-02-14
Publication date: 2023-06-27

Abstract

The disclosure provides a service pushing and calibration parameter obtaining method and device, wherein the service pushing method comprises the following steps: acquiring calibration parameters corresponding to the target class clusters; the calibration parameters corresponding to the target class cluster are determined based on the statistic value of the business index of the first sample business data of the target class cluster and the preset value of the business index of the second sample business data of the target class cluster; the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on the first sample service data of the plurality of class clusters; calibrating service indexes of target service data of the target class cluster based on calibration parameters corresponding to the target class cluster; the target service data and the second sample service data are service data of the same service scene, and at least part of the service data in the first sample service data and the second sample service data are service data of different service scenes; and pushing target business data to the target user based on the calibrated business indexes.

Description

Service pushing and calibration parameter obtaining method and device

Technical Field

The disclosure relates to the technical field of computer software, and in particular relates to a method and a device for acquiring service pushing and calibration parameters.

Background

When pushing service data, the service index of the service data is estimated first, and the service data with a larger estimated value is pushed to the user so as to achieve a certain service target. In order to improve the accuracy of the estimated value of the traffic indicator, a calibration technique is generally used to calibrate the estimated value of the traffic indicator, so that the estimated value is closer to the accurate value of the user behavior under the condition of keeping the order. The calibration technique relies on a large amount of sample traffic data, however, in some cases the sample traffic data is sparse, resulting in a low confidence in the calibration results, which affects the push effect of the traffic data.

Disclosure of Invention

In a first aspect of an embodiment of the present disclosure, a service pushing method is provided, where the method includes: acquiring calibration parameters corresponding to the target class clusters; the calibration parameters corresponding to the target class cluster are determined based on the statistic value of the service index of the first sample service data of the target class cluster and the pre-estimated value of the service index of the second sample service data of the target class cluster; the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on the first sample service data of a plurality of class clusters; calibrating service indexes of target service data of the target class cluster based on the calibration parameters corresponding to the target class cluster; the target service data and the second sample service data are service data of the same service scene, and at least part of the service data in the first sample service data and the second sample service data are service data of different service scenes; and pushing the target business data to a target user based on the calibrated business indexes.

In a second aspect of the embodiments of the present disclosure, there is provided a method for acquiring calibration parameters, the method including: acquiring a statistic value of a service index of first sample service data of a target class cluster; acquiring an estimated value of a service index of second sample service data of the target class cluster, wherein the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on first sample service data of a plurality of class clusters; determining a calibration parameter corresponding to the target class cluster based on the statistic value and the preset value; the calibration parameters corresponding to the target class cluster are used for calibrating service indexes of target service data of the target class cluster, the target service data and the second sample service data are service data of the same service scene, and at least part of service data in the first sample service data and the second sample service data are service data of different service scenes.

In a third aspect of the embodiments of the present disclosure, a service pushing apparatus is provided, where the apparatus includes: the first acquisition module is used for acquiring calibration parameters corresponding to the target class cluster; the calibration parameters corresponding to the target class cluster are determined based on the statistic value of the service index of the first sample service data of the target class cluster and the pre-estimated value of the service index of the second sample service data of the target class cluster; the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on the first sample service data of a plurality of class clusters; the calibration module is used for calibrating the service index of the target service data of the target class cluster based on the calibration parameters corresponding to the target class cluster; the target service data and the second sample service data are service data of the same service scene, and at least part of the service data in the first sample service data and the second sample service data are service data of different service scenes; and the pushing module is used for pushing the target business data to the target user based on the calibrated business indexes.

In a fourth aspect of embodiments of the present disclosure, there is provided an acquisition apparatus for calibration parameters, the apparatus including: the second acquisition module is used for acquiring the statistical value of the service index of the first sample service data of the target class cluster; a third obtaining module, configured to obtain an estimated value of a service index of second sample service data of the target class cluster, where the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on first sample service data of a plurality of class clusters; the determining module is used for determining the calibration parameters corresponding to the target class cluster based on the statistic value and the preset value; the calibration parameters corresponding to the target class cluster are used for calibrating service indexes of target service data of the target class cluster, the target service data and the second sample service data are service data of the same service scene, and at least part of service data in the first sample service data and the second sample service data are service data of different service scenes.

A fifth aspect of embodiments of the present disclosure provides a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the method of the first aspect described above.

A sixth aspect of the disclosed embodiments provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of the first aspect described above when executing the program.

According to the embodiment of the disclosure, a two-section calibration mode is adopted, the generation of the calibration tree and the output of the calibration parameters are split, on one hand, in the generation process of the calibration tree, the calibration tree is generated based on the first sample service data, and as at least part of service data in the first sample service data and the second sample service data are service data of different service scenes, namely, the service data of other service scenes are migrated to the service scene to which the second sample service data belongs to generate the calibration tree, the number of the sample data is increased, and the confidence of the calibration parameters is improved; on the other hand, in the process of generating the calibration parameters, the calibration tree is utilized to divide the second sample service data into clusters, and the estimated value of the service index of the second sample service data of the target cluster is obtained to generate the calibration parameters, so that the problem that the estimated value of the service index in the service scene to which the second sample service data belongs does not exist in the first sample service data of other service scenes is solved. In summary, the scheme of the embodiment of the disclosure can improve the confidence coefficient of the calibration parameter of the target scene, thereby improving the pushing effect of the service data.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings that are required for the embodiments or the description of the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present disclosure, and other drawings may be obtained according to these drawings without inventive effort for a person of ordinary skill in the art.

Fig. 1 is a flowchart of a service push method according to an embodiment of the present disclosure.

FIG. 2 is a schematic diagram of a calibration tree according to an embodiment of the present disclosure.

Fig. 3 is a general flow chart of an embodiment of the present disclosure.

Fig. 4 is a flowchart of a method for acquiring calibration parameters according to an embodiment of the present disclosure.

Fig. 5 is a block diagram of a service pushing device according to an embodiment of the disclosure.

Fig. 6 is a block diagram of an acquisition device of calibration parameters according to an embodiment of the present disclosure.

FIG. 7 is a schematic diagram of a computer device in one embodiment of the disclosure.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims.

The terminology used in the present disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. As used in this disclosure and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or all possible combinations of one or more of the associated listed items.

It should be understood that although the terms first, second, third, etc. may be used in this disclosure to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present disclosure. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "responsive to a determination", depending on the context.

User information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, presented data, etc.) referred to in this disclosure are both user-authorized or fully authorized information and data by parties, and the collection, use and processing of relevant data requires compliance with relevant laws and regulations and standards of relevant countries and regions, and is provided with corresponding operation portals for user selection of authorization or denial.

When pushing service data, the service index of the service data is estimated first, and the service data with a larger estimated value is pushed to the user. For example, in practical application, the service data may be an advertisement of a merchant or commodity to be pushed, and the service index may be a click rate or a conversion rate. Taking the click rate as an example, the click rate of a plurality of advertisements to be pushed can be estimated to obtain the predicted value of the click rate of each advertisement by the user, and then each advertisement is pushed to the user in sequence according to the sequence from the high click rate to the low click rate. In the above scenario, the predicted value of the click rate is used as a core technical support for intelligent bidding of advertisements and some sort mechanisms, so that the advertisement rendering efficiency is directly affected, and therefore, how to ensure the confidence of the predicted value of the click rate is important.

To increase the confidence level of the estimated value of the traffic indicator, a calibration technique is generally used to calibrate the estimated value of the traffic indicator. However, the calibration technique relies on a large amount of sample service data, and in some situations (such as small flow, low conversion rate, a pre-estimated model in a cold start period, etc.), the sample service data is sparse, so that the confidence of the calibration result is low, thereby affecting the pushing effect of the service data, and making it difficult for the service data pushed to the user to reach the expected click rate or conversion rate. If the service data of other service scenes are to be migrated to the target service scene to obtain the calibration parameters of the target service scene, there is a problem that the service data of other service scenes do not have the predicted value of the target service scene, so that it is difficult to simply migrate the service data of other service scenes to the target service scene.

Based on this, an embodiment of the present disclosure provides a service pushing method, referring to fig. 1, the method includes:

step S102: acquiring calibration parameters corresponding to the target class clusters; the calibration parameters corresponding to the target class cluster are determined based on the statistic value of the service index of the first sample service data of the target class cluster and the pre-estimated value of the service index of the second sample service data of the target class cluster; the target class cluster to which the target service data belong is determined based on a calibration tree, and the calibration tree is generated based on first sample service data of a plurality of class clusters;

step S104: calibrating service indexes of target service data of the target class cluster based on the calibration parameters corresponding to the target class cluster; the target service data and the second sample service data are service data of the same service scene, and at least part of the service data in the first sample service data and the second sample service data are service data of different service scenes;

step S106: and pushing the target business data to a target user based on the calibrated business indexes.

In step S102, the first sample service data and the second sample service data may be service data collected historically, for example, service data collected in the past 1 hour, the past 1 day, the past 1 week, or the past 1 month, or service data collected in some preset period of time in the past, for example, service data collected 11:00 to 13:00 before 1 day, data collected on the weekend of the last week, or the like. Taking the first sample service data as an example, in some application scenarios, a client may send a service request to a server, which may return service data to the client in response to the service request. The user may perform certain operations on the service data. The operation of the user on the service data can be recorded in the log, and the first sample service data can be obtained by pulling the data in the log.

For example, in a scenario where merchandise (e.g., clothing, dishes, toys, etc.) is pushed to a user, the service request sent by the client to the server may be a merchandise acquisition request. The server may return a list of pushed items (i.e., the business data described above) to the client in response to the item acquisition request. The user may perform at least one of browsing, clicking, joining a shopping cart, collecting, purchasing, commenting on items in the list. The user's operations on the merchandise may be recorded in a log. By pulling the log, commodity information of each commodity pushed to the user and operation information of the user on the commodity can be obtained, and commodity information of each commodity pushed to the user is used as first sample service data.

The second sample service data is obtained in a similar manner to the first sample service data, and will not be described here again. It will be appreciated that the above is only one exemplary manner of obtaining sample traffic data and is not intended to limit the present disclosure. In addition to the above-listed manners, other manners of acquiring the first sample traffic data and the second sample traffic data may be employed.

At least part of the service data in the first sample service data and the second sample service data are service data of different service scenes. For example, the first sample traffic data may include traffic data of a plurality of traffic scenes, and the second sample traffic data may include only traffic data of a target traffic scene of the plurality of scenes. Assuming that different cities are used as different service scenarios, the first service scenario may include a plurality of different cities such as Beijing, shanghai, hangzhou, guangzhou, and the like; accordingly, the first sample business data may include business data of a plurality of different cities such as Beijing, shanghai, hangzhou, guangzhou, and the like. And the second business scenario may include Beijing only; accordingly, the second sample service data may include only the service data of beijing. Alternatively, the first sample traffic data may include traffic data of a plurality of traffic scenes, and the second sample traffic data may include traffic data of a target traffic scene other than the plurality of scenes. For example, the first business scenario may include a plurality of different cities of Beijing, shanghai, hangzhou, guangzhou, and the like; accordingly, the first sample business data may include business data of a plurality of different cities such as Beijing, shanghai, hangzhou, guangzhou, and the like. And the second traffic scenario may comprise Shenzhen; accordingly, the second sample service data may include Shenzhen service data.

The above embodiments take different cities as examples of different service scenarios, and illustrate the relationship between the first sample service data and the second sample service data. It will be appreciated that the foregoing is merely illustrative, and in other embodiments, the service scenario may be differentiated according to conditions other than city, which is not described herein.

The first sample traffic data and the second sample traffic data may each be divided into a plurality of class clusters. Still taking the first sample service data as an example, the first sample service data may be divided into two different class clusters according to whether the first sample service data is acquired from an independent client (Application, APP), that is, the first sample service data acquired from an independent APP is divided into one class cluster, and the first sample service data acquired from other channels (e.g., a WeChat applet, a payment device front page, etc.) is divided into another class cluster. For the first sample service data obtained from the independent APP, class clusters may be further divided according to a traffic scenario (abbreviated as a first traffic scenario) of the first sample service data obtained from the independent APP, for example, the first sample service data recommended by the first traffic scenario as the first page is divided into one class cluster, and the first sample service data of the first traffic scenario as other traffic scenario is divided into another class cluster.

For the first sample service data obtained from other channels, the classification clusters may be further classified according to the category of the first sample service data, for example, the first sample service data of the food category is classified into one category cluster, and the first sample service data of other categories (for example, the category of clothing, toys, etc.) is classified into another category cluster. For the first sample service data of the food purpose, the first sample service data of the food purpose (abbreviated as a second flow scene) can be divided into different class clusters according to the flow scene of the first sample service data of the food purpose, for example, the first sample service data of which the second flow scene is "over-value preemption" is divided into one class cluster, and the first sample service data of which the second flow scene is other flow scene is divided into another class cluster.

For the first sample service data of other types, the unit types of the first sample service data may be divided into different class clusters, for example, the first sample service data of which the unit type is a store (i.e., the first sample service data is the service data of a store), may be divided into one class cluster, and the first sample service data of which the unit type is a commodity (i.e., the first sample service data is the service data of a single commodity), may be divided into another class cluster.

It is to be understood that the above is only one exemplary manner of classifying clusters and is not intended to limit the present disclosure. In practical applications, the first sample service data may be divided into a plurality of class clusters in other manners, which are not exemplified here. The class clusters included in the second sample service data may be the same as the class clusters included in the first sample service data, or may be a subset of the class clusters included in the first sample service data due to the sparse number of the second sample service data.

In some embodiments, the second sample traffic data may be clustered by a calibration tree. The calibration tree may be a classification and regression (Classification And Regression Tree, CART) tree, and the calibration tree may include a plurality of nodes, each node corresponding to a class cluster, and a class cluster corresponding to a child node of a node being a subset of the class cluster corresponding to the node.

Referring to fig. 2, a schematic diagram of a calibration tree of some embodiments is shown. Each circle represents a node, the node marked as Root represents a Root node, and the class cluster corresponding to the Root node can comprise the full amount of second sample service data. The square to which each non-leaf node is connected represents a division condition (also referred to as a feature) for further dividing the class cluster corresponding to the non-leaf node into sub-class clusters. For example, the square to which the root node is connected is "client", which indicates that the partition condition for further dividing the class cluster corresponding to the root node into the sub-class clusters is: and whether the second sample service data in the class cluster corresponding to the root node is acquired from the independent client or not. For another example, the square to which the right child node of the root node (i.e., the node identified as 2 in the figure) is connected is "category", which indicates that the division condition for further dividing the class cluster corresponding to the right child node of the root node into the sub-class clusters is: and the category to which the second sample service data in the category cluster corresponding to the right child node of the root node belongs.

One feature may correspond to a plurality of feature values, and each second sample service data in the same class of cluster may be divided into different class clusters according to the feature value of each second sample service data in the class of cluster. The determination of which class of clusters the second sample traffic data is divided into may be made by comparing the characteristic value of the second sample traffic data with a corresponding node split threshold. For example, if the characteristic value of the second sample service data is the same as the corresponding node splitting threshold value, the second sample service data is divided into a class cluster; and if the characteristic value of the second sample service data is different from the corresponding node splitting threshold value, dividing the second sample service data into another class cluster. For another example, for a value type of feature value and a node split threshold, if the feature value of the second sample traffic data is greater than or equal to the corresponding node split threshold, the second sample traffic data is partitioned into one class cluster; and if the characteristic value of the second sample service data is smaller than the corresponding node splitting threshold value, dividing the second sample service data into another class cluster.

In the example shown in fig. 2, the feature value corresponding to the feature of "client" may include "independent APP" and "other", and the node splitting threshold may be "independent APP". Therefore, if a certain piece of second sample service data in the class cluster corresponding to the root node is obtained from the independent client, the characteristic value of the second sample service data is the same as the corresponding node splitting threshold, so that the second sample service data can be divided into the class cluster corresponding to the left child node of the root node (namely, the node marked as 1 in the figure); if a certain piece of second sample service data in the class cluster corresponding to the root node is obtained from other channels, the characteristic value of the second sample service data is different from the corresponding node splitting threshold value, so that the second sample service data can be divided into the class clusters corresponding to the right child node of the root node. In this way, the second sample service data in the class cluster corresponding to the left child node of the root node is the service data obtained from the independent APP in the full amount of second sample service data, and the second sample service data in the class cluster corresponding to the right child node of the root node is the service data obtained from other channels in the full amount of second sample service data. That is, the class cluster corresponding to the left child node of the root node and the class cluster corresponding to the right child node of the root node are both subsets of the class cluster corresponding to the root node. Similarly, each piece of second sample service data in the class clusters corresponding to other nodes can be further divided into different class clusters.

In the embodiment, the calibration tree is adopted to divide the class clusters, the tree model has strong interpretation, the calibration according to the feature division granularity can be clearly known, in addition, the class clusters can be automatically divided by adopting the tree model, the confidence level is ensured through the node division threshold value, and the number of the class clusters is not required to be designated in advance. As shown in fig. 2, the calibration tree may be a binary tree. Binary trees have a deeper depth and cluster-like partitioning that is finer and more accurate than multi-way trees. In other embodiments, the calibration tree may be other tree models than a binary tree.

The calibration tree may be generated based on the first sample traffic data. Taking the example that the calibration tree is a binary tree, the following steps may be performed for each candidate node loop in the calibration tree starting from the root node: dividing first sample service data in candidate class clusters corresponding to the candidate nodes into a left sub-class cluster or a right sub-class cluster based on each candidate feature in a feature set and each candidate feature value in a feature value set corresponding to the candidate feature, and determining a distance between the left sub-class cluster and the right sub-class cluster based on a service index of the first sample service data of the left sub-class cluster and a service index of the first sample service data of the right sub-class cluster; if the distance between the left sub-cluster and the right sub-cluster divided based on at least one group of candidate features and candidate feature values is larger than a preset distance threshold, determining the candidate features and candidate feature values corresponding to the maximum distance as features and candidate feature values corresponding to the candidate nodes, adding left sub-nodes and right sub-nodes to the candidate nodes, and determining the left sub-nodes and the right sub-nodes as candidate nodes.

Wherein one or more features may be included in the feature set, and one feature may be selected from the feature set at a time as a candidate feature. Each feature in the feature set may correspond to a set of feature values that includes one or more feature values. Assume that the feature set is denoted as { F ₁ ,F ₂ }, wherein F ₁ And F ₂ Are all features in the feature set, assume feature F ₁ The corresponding set of eigenvalues is denoted as { V ₁₁ ,V ₁₂ Characteristic F ₂ The corresponding set of eigenvalues is denoted as { V ₂₁ ,V ₂₂ ,V ₂₃ }, wherein V is ₁₁ ,V ₁₂ ,V ₂₁ ,V ₂₂ ,V ₂₃ Are all eigenvalues. For one of the candidate nodes in the calibration tree, 5 sets of candidate features and candidate features may be selected from the feature set and the feature value setThe sign values are respectively: { F ₁ ,V ₁₁ }，{F ₁ ,V ₁₂ }，{F ₂ ,V ₂₁ }，{F ₂ ,V ₂₂ Sum { F } ₂ ,V ₂₃ }. In { F ] ₁ ,V ₁₁ For example, the set of candidate features and candidate feature values may be used to partition the first sample traffic data in the candidate class cluster corresponding to the candidate node into a left sub-class cluster or a right sub-class cluster based on the set of candidate features and candidate feature values, and then determine a distance between the left sub-class cluster and the right sub-class cluster. The same may be performed for the remaining 4 sets of candidate features and candidate feature values. Finally obtaining 5 distances respectively corresponding to different candidate features and combinations of candidate feature values, and assuming { F ₁ ,V ₁₁ The distance between the divided left and right sub-clusters is the largest distance among the above 5 distances.

F if at least one distance greater than the preset distance threshold exists in the 5 distances ₁ And V ₁₁ And respectively determining the characteristics and the candidate characteristic values corresponding to the candidate nodes, and adding a left child node and a right child node to the candidate nodes. Wherein the first sample service data in the class cluster corresponding to the left child node and the first sample service data in the class cluster corresponding to the right child node are based on { F } ₁ ,V ₁₁ And dividing the first sample service data in the class cluster corresponding to the candidate node. If there is no distance greater than a preset distance threshold value among the 5 distances, the candidate node is determined as a leaf node. If each candidate node is traversed, the calibration tree is generated.

In some embodiments, the distance between the left and right sub-category clusters may be determined based on: acquiring a first statistical value of a service index of first sample service data of a left sub-cluster; acquiring a second statistical value of a service index of the first sample service data of the right sub-cluster; the distance between the left and right sub-cluster is determined based on the absolute difference of the first and second statistics.

Taking the click rate as the service index as an example, assume that the number of the first sample service data in the left sub-cluster is m _L Clicked by the user in the left sub-clusterThe number of the first sample service data is n _L The first statistic may be denoted as CTR _DL ＝n _L /m _L . Assume that the number of the first sample service data in the right sub-cluster is m _R The number of the first sample service data clicked by the user in the right sub-cluster is n _R The second statistic may be noted as CTR _DR ＝n _R /m _R . The distance between the left and right sub-clusters can be noted as |CTR _DL -CTR _DR | a. The invention relates to a method for producing a fibre-reinforced plastic composite. The characteristics corresponding to the candidate nodes and the candidate characteristic values satisfy the following conditions:

wherein F is _i For the ith feature in the feature set, V _ij Is F _i And j-th eigenvalue in the corresponding eigenvalue set. In the above embodiment, the features in the feature set and the feature values in the feature value set may be defined by themselves according to actual needs, so as to customize the stage division criteria. Therefore, the calibration type clusters can be automatically divided, and the multi-dimensional calibration effect and flexibility are improved.

In general, the calibration model is obtained by training and optimizing on the same batch of sample service data, that is, the calibration parameters are obtained by training according to the statistic value and the pre-estimated value of the service indexes of the sample service data of the same batch and the same class of clusters. However, because the sample service data of the small-flow scene is sparse, a calibration model with higher confidence coefficient cannot be obtained through training. Through data analysis, the inventor finds that although service scenes are different or the estimated models are different (for example, the estimated model of new online is different from the estimated model adopted at present), the distribution of sample service data is similar under a certain granularity (such as the same category), which provides possibility for migration multiplexing of the sample service data among the service scenes. Therefore, in order to improve the confidence, the sample service data of other similar service scenes are expanded in to obtain the predicted value of the target service scene. However, there is a problem that the sample service data of other service scenes is not predicted value of the target scene, for example, only the predicted value of the click rate of the user in the Beijing area on the sample service data of a certain class of clusters can be obtained according to the sample service data in the Beijing area, but the predicted value of the click rate of the user in the Hangzhou area on the sample service data of a certain class of clusters cannot be obtained. In order to solve the problem, the method adopts a two-section calibration mode to split the generation of the calibration tree and the output of the calibration parameters, firstly uses the first sample service data of a plurality of service scenes to generate the calibration tree, then uses the calibration tree to divide the class clusters of the second sample service data (the second sample service data can have no conversion or clicking behaviors and only needs to have model pre-estimation values) of the target service scene, and finally outputs the calibration parameters of the corresponding class clusters according to the statistic value and the pre-estimation value of each class cluster. Under the condition that sample service data of a target service scene is sparse, the accuracy and the confidence of the estimated result can be weighed by the mode.

It should be noted that, in the related art, a root mean square error (Root Mean Square Error, RMSE) index of a ratio of an estimated value and a statistical value of a traffic index of a different type cluster is generally used to evaluate a cluster-like division effect, and the division target is to make the RMSE index of the ratio of the estimated value and the statistical value of the traffic index of the different type cluster as small as possible. However, in the embodiment of the present disclosure, since the first sample service data of other service scenarios is migrated, and the first sample service data has no predicted value of the target service scenario, it is not possible to simply perform cluster classification according to RMSE indexes. Different from the related art, the embodiment of the disclosure adopts the absolute value of the difference of the statistical mean values of the clusters to measure the distance between two clusters, and the larger the absolute value is, the larger the cluster distance is, so that the optimal characteristic and characteristic value are obtained. The method is adopted on one hand because the calibration is carried out on the basis of the cluster statistical mean value, and the larger the cluster statistical mean value is, the larger the distribution difference of the sample service data in the two clusters is, and on the other hand, the problem that other service scenes have no predicted value of the target service scene is avoided.

The first sample traffic data of the target class cluster may be traffic data belonging to the target class cluster in the first sample traffic data. For example, assuming that the first sample service data includes service data of Beijing, shanghai, guangzhou and Hangzhou, the first sample service data is divided into three kinds of clusters of toys, food and clothes according to categories, and the class cluster corresponding to the food category is a target class cluster, the first sample service data of the target class cluster may include service data belonging to the food category in service data of Beijing, service data belonging to the food category in service data of Shanghai, service data belonging to the food category in service data of Guangzhou, and service data belonging to the food category in service data of Hangzhou. Similarly, the second sample service data of the target class cluster may be service data belonging to the target class cluster in the second sample service data, which is not described herein for brevity.

Each cluster can correspond to a calibration parameter for calibrating the traffic index of the target traffic data of the cluster. For example, in the above class cluster division manner, the home page recommendation, the over-value purchase, the store, and the commodity respectively correspond to one class cluster, and each of these class clusters may correspond to one calibration parameter, and the calibration parameters corresponding to different class clusters may be the same or different. In addition, since the cate category and the other category also correspond to one category cluster, the category cluster corresponding to the cate category and the category cluster corresponding to the other category may also correspond to one calibration parameter.

The target class cluster may be any one of the respective class clusters included in the first sample service data. In practical application, calibration parameters corresponding to each class cluster can be predetermined, and under the condition that target service data is received, the class cluster to which the target service data belongs is determined as the target class cluster, so that the calibration parameters corresponding to the target class cluster are obtained.

The calibration parameter corresponding to the target class cluster may be determined based on a statistical value of a traffic index of the first sample traffic data of the target class cluster and a predicted value of a traffic index of the second sample traffic data of the target class cluster. The business index may be a click through Rate (Click Through Rate, CTR) or Conversion Rate (CVR) parameter. The label of each first sample service data of the target class cluster can be obtained, and the statistical value of the click rate of the first sample service data of the target class cluster is determined according to the label of each first sample service data of the target class cluster. Wherein, the label corresponding to the first sample service data is related to the service index of the first sample service data.

In some embodiments, the business index is a click rate, and the tag of the first sample business data is used to characterize whether the first sample business data was clicked by the user. In this case, the click rate of the first sample service data in the target class cluster may be determined according to the label of each first sample service data in the target class cluster, and the ratio between the click rate and the total number of the first sample service data in the target class cluster may be determined as the statistical value of the click rate of the first sample service data in the target class cluster.

In other embodiments, the business index is conversion rate and the label of the first sample business data is used to characterize whether the first sample business data achieves user conversion. In this case, the conversion amount of the first sample service data in the target class cluster may be determined according to the label of each of the first sample service data in the target class cluster, and the ratio between the conversion amount and the total number of the first sample service data in the target class cluster may be determined as the statistical value of the conversion rate of the first sample service data of the target class cluster.

In the case where the first sample service data and the corresponding operation information thereof are recorded in the log, the operation information of the user on the first sample service data may be acquired from the log, and the tag of the first sample service data may be generated based on the operation information. Still taking the example that the service index is the click rate, if the operation information of the user on the first sample service data is recorded in the log and includes a click operation, the tag information of the first sample service data may indicate that the first sample service data is clicked by the user, for example, indicated by the identification information "1"; if the operation information of the user on the first sample service data recorded in the log does not include a click operation, the tag information of the first sample service data may indicate that the first sample service data has not been clicked by the user, for example, indicated by the identification information "0".

The estimated value of the service index can be obtained through a pre-trained estimated model, wherein the estimated model can be obtained through training based on first sample service data of a plurality of class clusters and labels corresponding to the first sample service data. Further, the predictive model may be trained jointly based on the first sample service data, the tag corresponding to the first sample service data, and the user characteristics of the user to whom the first sample service data is pushed. User characteristics may include, but are not limited to, at least one of age, gender, occupation, location, etc. of the user, and these user characteristics may be obtained from a client used by the user with authorization of the user.

After the statistical value of the service index of the first sample service data of the target class cluster and the predicted value of the service index of the second sample service data of the target class cluster are obtained, the calibration parameter corresponding to the target class cluster can be determined based on the ratio between the statistical value of the service index of the first sample service data of the target class cluster and the predicted value of the service index of the second sample service data of the target class cluster. Still taking the click rate as an example of the business index, assuming that the statistic value is denoted as ctr and the pre-determined value is denoted as pctr, the calibration parameter alpha may be denoted as alpha=ctr/pctr.

In step S104, the traffic index of the target traffic data of the target class cluster may be calibrated based on the calibration parameter acquired in step S102. The target service data and the second sample service data are service data of the same service scene. The target service data may be service data of a target service scenario acquired in response to a real-time service request. In some embodiments, the estimated value of the traffic index of the target traffic data may be obtained through the estimation model described in the above embodiments, and the estimated value is multiplied by the calibration parameter alpha, so as to obtain the traffic index after calibration of the target traffic data. It will be appreciated that in the above embodiment, the calibration parameter alpha may also be denoted as alpha=pctr/ctr. In this case, the estimated value of the service index of the target service data may be obtained through the estimation model, and the estimated value is divided by the calibration parameter alpha, thereby obtaining the service index after the calibration of the target service data.

In step S106, the target service data may be pushed to the target user based on the calibrated service index acquired in step S104. For example, the multi-label business data may be pushed to the user in the order of the business index after the calibration of the multi-label business data from large to small. For another example, the target service data may be pushed to the target user when the service index after the calibration of the target service data is greater than or equal to the preset index threshold, and not pushed to the target user when the service index after the calibration of the target service data is less than the preset index threshold.

In some embodiments, calibration parameters corresponding to each class cluster may also be stored in an index corresponding to the calibration tree. After receiving the target service data, the calibration parameters corresponding to the target class cluster to which the target service data belongs can be obtained from the index.

Specifically, for one node X in the calibration tree, assuming that the node X corresponds to the class cluster Y, the index may store a feature value of each node on the path from the root node to the node X and a calibration parameter corresponding to the class cluster Y. Wherein one path in the calibration tree comprises a set of nodes { Node } in the calibration tree ₁ ，Node ₂ ，……，Node _r Node in the group of nodes }, and _p for Node _p+1 Is 1.ltoreq.p<r. For example, in the embodiment shown in FIG. 2, the root node, node identified as 1, and node identified as 3 make up one path (denoted path 1), and the root node, node identified as 2, node identified as 4, and node identified as 5 make up another path (denoted path 2). Thus, for path 1, the characteristic value of the root node, the characteristic value of the node identified as 1, the characteristic value of the node identified as 3, and the calibration parameter corresponding to the class cluster corresponding to the node identified as 3 may be recorded in the index. Since the class cluster corresponding to the root node generally comprises the total amount of service data, the characteristic value of the root node can be ignored. The feature value and the calibration parameter may be stored in the form of key value pairs, where a key (key) is a feature value of each node on the path where the node X is located, and a value (value) is the calibration parameter, where a storage mode is key: value.

Assuming that the calibration parameter is 0.5, the storage manner of the calibration parameter corresponding to the class cluster corresponding to the node identified as 3 in the index may be expressed as follows: independent app_home recommendation: 0.5. under the condition that the depth of the calibration tree is relatively large, by establishing the index, time delay generated in the process of searching the calibration tree can be effectively reduced, and the efficiency of acquiring the calibration parameters is improved.

Referring to fig. 3, a schematic diagram of the overall flow of an embodiment of the present disclosure is shown. As shown in fig. 3, an embodiment of the present disclosure may include the steps of:

step S302: sample service data is obtained from the log, including first sample service data of a plurality of service scenes (abbreviated as full scene samples) and second sample service data of a target service scene (abbreviated as target scene samples).

Step S304: a calibration tree is generated based on the full field Jing Yangben and the target scene samples are clustered based on the calibration tree.

Step S306: acquiring an estimated value of a service index of a target scene sample of a target class cluster (short class cluster estimated value) and a statistical value of a service index of a full scene sample of the target class cluster (short class cluster statistical value), and calculating a calibration parameter corresponding to the target class cluster based on the class cluster estimated value and the class cluster statistical value.

Step S308: and carrying out online calibration on the service index of the target service data corresponding to the target class cluster based on the calibration parameter corresponding to the target class cluster.

Step S310: and pushing the target service data to the target user based on the calibrated service index.

Here, steps S302 to S306 may be performed in advance in the training phase, and steps S308 and S310 may be performed in real time during the business process. That is, the calibration parameters may be acquired in advance in the training phase, and in the business processing phase, only the calibration parameters need to be acquired, and the steps S308 and S310 need not be repeatedly performed, so that the steps S302 to S306 do not need to be repeatedly performed.

Referring to fig. 4, an embodiment of the present disclosure further provides a method for obtaining a calibration parameter, where the method includes:

step S402: acquiring a statistic value of a service index of first sample service data of a target class cluster;

step S404: acquiring an estimated value of a service index of second sample service data of the target class cluster, wherein the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on first sample service data of a plurality of class clusters;

step S406: determining a calibration parameter corresponding to the target class cluster based on the statistic value and the preset value; the calibration parameters corresponding to the target class cluster are used for calibrating service indexes of target service data of the target class cluster, the target service data and the second sample service data are service data of the same service scene, and at least part of service data in the first sample service data and the second sample service data are service data of different service scenes.

The specific embodiments of each step in the method for obtaining the calibration parameter may refer to the embodiments of the service pushing method described above, and will not be described herein.

It will be appreciated that the solutions described in the above embodiments can be freely combined to obtain new solutions without conflicts, and for reasons of brevity, will not be described further herein.

The following are device embodiments of the present disclosure that may be used to perform method embodiments of the present disclosure. For details not disclosed in the embodiments of the apparatus of the present disclosure, please refer to the embodiments of the method of the present disclosure.

Fig. 5 is a block diagram of a service pushing device according to an embodiment of the disclosure, the device including:

a first obtaining module 502, configured to obtain a calibration parameter corresponding to a target cluster; the calibration parameters corresponding to the target class cluster are determined based on the statistic value of the service index of the first sample service data of the target class cluster and the pre-estimated value of the service index of the second sample service data of the target class cluster; the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on the first sample service data of a plurality of class clusters;

a calibration module 504, configured to calibrate a traffic indicator of target traffic data of the target class cluster based on a calibration parameter corresponding to the target class cluster; the target service data and the second sample service data are service data of the same service scene, and at least part of the service data in the first sample service data and the second sample service data are service data of different service scenes;

And a pushing module 506, configured to push the target service data to the target user based on the calibrated service indicator.

Fig. 6 is a block diagram of an apparatus for obtaining calibration parameters according to an embodiment of the present disclosure, the apparatus comprising:

a second obtaining module 602, configured to obtain a statistical value of a service index of the first sample service data of the target class cluster;

a third obtaining module 604, configured to obtain an estimated value of a service index of second sample service data of the target class cluster, where the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on first sample service data of a plurality of class clusters;

a determining module 606, configured to determine a calibration parameter corresponding to the target cluster based on the statistics and the preset value; the calibration parameters corresponding to the target class cluster are used for calibrating service indexes of target service data of the target class cluster, the target service data and the second sample service data are service data of the same service scene, and at least part of service data in the first sample service data and the second sample service data are service data of different service scenes.

In some embodiments, functions or modules included in an apparatus provided by the embodiments of the present disclosure may be used to perform a method described in the foregoing method embodiments, and specific implementations thereof may refer to descriptions of the foregoing method embodiments, which are not repeated herein for brevity.

The embodiments of the present disclosure also provide a computer device at least including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of the preceding embodiments when executing the program.

FIG. 7 illustrates a more specific hardware architecture diagram of a computing device provided by embodiments of the present description, which may include: a processor 702, a memory 704, an input/output interface 706, a communication interface 708, and a bus 710. Wherein the processor 702, the memory 704, the input/output interface 706 and the communication interface 708 enable communication connections between each other within the device via a bus 710.

The processor 702 may be implemented by a general-purpose CPU (Central Processing Unit ), microprocessor, application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, etc. for executing relevant programs to implement the technical solutions provided in the embodiments of the present disclosure. The processor 702 may also include a graphics card, which may be an Nvidia titanium X graphics card, a 1080Ti graphics card, or the like.

The Memory 704 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory ), static storage device, dynamic storage device, or the like. Memory 704 may store an operating system and other application programs, and when the embodiments of the present specification are implemented in software or firmware, the associated program code is stored in memory 704 and executed by processor 702.

The input/output interface 706 is used to connect with an input/output module to realize information input and output. The input/output module may be configured as a component in a device (not shown) or may be external to the device to provide corresponding functionality. Wherein the input devices may include a keyboard, mouse, touch screen, microphone, various types of sensors, etc., and the output devices may include a display, speaker, vibrator, indicator lights, etc.

The communication interface 708 is used to connect communication modules (not shown) to enable communication interactions of the device with other devices. The communication module may implement communication through a wired manner (such as USB, network cable, etc.), or may implement communication through a wireless manner (such as mobile network, WIFI, bluetooth, etc.).

Bus 710 includes a path to transfer information between components of the device (e.g., processor 702, memory 704, input/output interface 706, and communication interface 708).

It should be noted that although the above-described device only shows the processor 702, the memory 704, the input/output interface 706, the communication interface 708, and the bus 710, in a specific implementation, the device may also include other components necessary to achieve proper operation. Furthermore, it will be understood by those skilled in the art that the above-described apparatus may include only the components necessary to implement the embodiments of the present description, and not all the components shown in the drawings.

The disclosed embodiments also provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the method of any of the previous embodiments.

Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.

From the foregoing description of embodiments, it will be apparent to those skilled in the art that the present embodiments may be implemented in software plus a necessary general purpose hardware platform. Based on such understanding, the technical solutions of the embodiments of the present specification may be embodied in essence or what contributes to the prior art in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the embodiments or some parts of the embodiments of the present specification.

The system, apparatus, module or unit set forth in the above embodiments may be implemented in particular by a computer chip or entity, or by a product having a certain function. A typical implementation device is a computer, which may be in the form of a personal computer, a laptop computer, a cellular telephone, an image capture device telephone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for the device embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference is made to the description of the method embodiments for relevant points. The apparatus embodiments described above are merely illustrative, in which the modules illustrated as separate components may or may not be physically separate, and the functions of the modules may be implemented in the same piece or pieces of software and/or hardware when implementing the embodiments of the present disclosure. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.

The foregoing is merely a specific implementation of the embodiments of this disclosure, and it should be noted that, for a person skilled in the art, several improvements and modifications may be made without departing from the principles of the embodiments of this disclosure, and these improvements and modifications should also be considered as protective scope of the embodiments of this disclosure.

Claims

1. A method of traffic push, the method comprising:

acquiring calibration parameters corresponding to the target class clusters; the calibration parameters corresponding to the target class cluster are determined based on the statistic value of the service index of the first sample service data of the target class cluster and the pre-estimated value of the service index of the second sample service data of the target class cluster; the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on the first sample service data of a plurality of class clusters;

calibrating service indexes of target service data of the target class cluster based on the calibration parameters corresponding to the target class cluster; the target service data and the second sample service data are service data of the same service scene, and at least part of the service data in the first sample service data and the second sample service data are service data of different service scenes;

and pushing the target business data to a target user based on the calibrated business indexes.

2. The method of claim 1, the first sample traffic data comprising traffic data of a plurality of traffic scenarios, the second sample traffic data comprising traffic data of a target traffic scenario of the plurality of scenarios.

3. The method of claim 1, the calibration tree comprising a plurality of nodes, each node corresponding to a cluster of classes; the calibration tree is generated by:

the following steps are cyclically performed for each candidate node in the calibration tree starting from the root node:

dividing first sample service data in candidate class clusters corresponding to the candidate nodes into a left sub-class cluster or a right sub-class cluster based on each candidate feature in a feature set and each candidate feature value in a feature value set corresponding to the candidate feature, and determining a distance between the left sub-class cluster and the right sub-class cluster based on a service index of the first sample service data of the left sub-class cluster and a service index of the first sample service data of the right sub-class cluster;

if the distance between the left sub-cluster and the right sub-cluster divided based on at least one group of candidate features and candidate feature values is larger than a preset distance threshold, determining the candidate features and candidate feature values corresponding to the maximum distance as features and candidate feature values corresponding to the candidate nodes, adding left sub-nodes and right sub-nodes to the candidate nodes, and determining the left sub-nodes and the right sub-nodes as candidate nodes.

4. The method of claim 3, the determining the distance between the left and right sub-cluster based on the traffic index of the first sample traffic data of the left sub-cluster and the traffic index of the first sample traffic data of the right sub-cluster comprising:

acquiring a first statistical value of a service index of the first sample service data of the left sub-cluster;

acquiring a second statistical value of a service index of the first sample service data of the right sub-cluster;

a distance between the left and right sub-cluster is determined based on an absolute value of a difference between the first and second statistics.

5. The method according to any one of claims 1 to 4, wherein the calibration parameter corresponding to the target class cluster is determined based on a ratio between a statistical value of traffic indexes of the first sample traffic data of the target class cluster and a predicted value of traffic indexes of the second sample traffic data of the target class cluster.

6. A method of obtaining calibration parameters, the method comprising:

acquiring a statistic value of a service index of first sample service data of a target class cluster;

acquiring an estimated value of a service index of second sample service data of the target class cluster, wherein the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on first sample service data of a plurality of class clusters;

Determining a calibration parameter corresponding to the target class cluster based on the statistic value and the preset value; the calibration parameters corresponding to the target class cluster are used for calibrating service indexes of target service data of the target class cluster, the target service data and the second sample service data are service data of the same service scene, and at least part of service data in the first sample service data and the second sample service data are service data of different service scenes.

7. A traffic pushing device, the device comprising:

the first acquisition module is used for acquiring calibration parameters corresponding to the target class cluster; the calibration parameters corresponding to the target class cluster are determined based on the statistic value of the service index of the first sample service data of the target class cluster and the pre-estimated value of the service index of the second sample service data of the target class cluster; the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on the first sample service data of a plurality of class clusters;

the calibration module is used for calibrating the service index of the target service data of the target class cluster based on the calibration parameters corresponding to the target class cluster; the target service data and the second sample service data are service data of the same service scene, and at least part of the service data in the first sample service data and the second sample service data are service data of different service scenes;

And the pushing module is used for pushing the target business data to the target user based on the calibrated business indexes.

8. An acquisition device of calibration parameters, the device comprising:

the second acquisition module is used for acquiring the statistical value of the service index of the first sample service data of the target class cluster;

a third obtaining module, configured to obtain an estimated value of a service index of second sample service data of the target class cluster, where the class cluster to which the second sample service data belongs is determined based on a calibration tree, and the calibration tree is generated based on first sample service data of a plurality of class clusters;

the determining module is used for determining the calibration parameters corresponding to the target class cluster based on the statistic value and the preset value; the calibration parameters corresponding to the target class cluster are used for calibrating service indexes of target service data of the target class cluster, the target service data and the second sample service data are service data of the same service scene, and at least part of service data in the first sample service data and the second sample service data are service data of different service scenes.

9. A computer readable storage medium having stored thereon a computer program which when executed by a processor implements the method of any of claims 1 to 6.

10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of any one of claims 1 to 6 when the program is executed.