CN111768244A - Advertisement delivery recommendation method and device - Google Patents

Advertisement delivery recommendation method and device Download PDF

Info

Publication number
CN111768244A
CN111768244A CN202010622122.XA CN202010622122A CN111768244A CN 111768244 A CN111768244 A CN 111768244A CN 202010622122 A CN202010622122 A CN 202010622122A CN 111768244 A CN111768244 A CN 111768244A
Authority
CN
China
Prior art keywords
target object
training
determining
click
user
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010622122.XA
Other languages
Chinese (zh)
Inventor
蔚静
韩海燕
卢道和
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
WeBank Co Ltd
Original Assignee
WeBank Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by WeBank Co Ltd filed Critical WeBank Co Ltd
Priority to CN202010622122.XA priority Critical patent/CN111768244A/en
Publication of CN111768244A publication Critical patent/CN111768244A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0255—Targeted advertisements based on user history
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/24323—Tree-organised classifiers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/08—Insurance

Abstract

The invention relates to a recommendation method and a device for advertisement putting, which comprises the following steps: determining a first candidate associated tag of a target object according to the corresponding relation between the object tag and the associated tag of the target object; determining the similarity between a target object and historical objects, and determining the similar objects of the target object from all the historical objects according to the similarity; selecting the actual associated label of the similar object as a second candidate associated label of the target object; inputting the candidate associated labels and the feature data of the target object into a click prediction model aiming at each candidate associated label of the target object to obtain the predicted click probability of the candidate associated labels to the target object; determining recommended associated labels from all candidate associated labels according to the predicted click probability; and determining an advertisement putting strategy of the target object according to the recommendation association label. The accuracy and pertinence of insurance product putting are increased, and the insurance product putting conversion rate is improved.

Description

Advertisement delivery recommendation method and device
Technical Field
The invention relates to the field of big data processing of science and technology finance (Fintech), in particular to a recommendation method and device for advertisement putting.
Background
With the continuous development of financial technologies, especially internet technology and finance, more and more technologies (such as distributed, big data, artificial intelligence, etc.) are applied in the financial field, but the financial industry also puts higher demands on the technologies.
Insurance products are a complex of tangible products and intangible services offered by insurance companies to the market. The insurance product refers to a financial tool which is created by an insurance company and can be selected by a client to trade in an insurance market in a narrow sense; broadly, it refers to all products and services that an insurance company offers to the market and can be acquired, utilized or consumed by customers, all of which fall within the category of insurance product services. Advertising of insurance products is required in order to promote or provide services to the insurance products.
In the existing method for releasing insurance products, generally, business personnel select a proper channel matched with the insurance products to release the insurance products according to the characteristics of the insurance products, and the method has strong subjectivity, low release accuracy and pertinence and can not effectively improve the release conversion rate of the insurance products.
Disclosure of Invention
The application provides a recommendation method and device for advertisement putting, which are used for increasing the accuracy and pertinence of putting insurance products and improving the putting conversion rate of the insurance products.
The advertisement delivery recommendation method provided by the embodiment of the invention comprises the following steps:
determining a first candidate associated tag of a target object according to the corresponding relation between the object tag and the associated tag of the target object;
determining the similarity between a target object and historical objects, and determining the similar objects of the target object from all the historical objects according to the similarity;
selecting the actual associated label of the similar object as a second candidate associated label of the target object;
inputting the candidate associated labels and the feature data of the target object into a click prediction model aiming at each candidate associated label of the target object to obtain the click probability of the candidate associated labels on the target object; the click prediction model is trained according to the characteristic data of the training sample and the actual click result to obtain corresponding model parameters;
determining recommended associated labels from all candidate associated labels according to the click probability;
and determining an advertisement putting strategy of the target object according to the recommendation association label.
In an optional embodiment, before determining the first candidate associated tag of the target object according to the correspondence between the object tag and the associated tag of the target object, the method further includes:
determining a preliminary associated tag of the historical object according to the characteristic data of the historical object;
acquiring user behavior data in the historical object putting process;
and modifying the preliminary association tag of the historical object according to the user behavior data to obtain an actual association tag of the historical object.
In an alternative embodiment, the click prediction model is trained according to the following:
acquiring a training sample, wherein the training sample comprises characteristic data of a training object, characteristic data of a training user and an actual click result of the training user on the training object;
inputting the feature data of the training object and the feature data of the training user into the click prediction model to obtain a predicted click result of the training user for the training object;
determining the predicted click probability of the training object according to the predicted click results of all training users aiming at the training object;
determining the actual click rate of the training object according to the actual click result of the training user on the training object;
and calculating a loss function according to the actual click rate and the predicted click probability, and determining a parameter corresponding to the click prediction model when the loss function is smaller than a preset threshold value.
In an alternative embodiment, the click prediction model comprises a GBDT gradient boosting iterative decision tree model and an LR logistic regression model;
the inputting the feature data of the training object and the feature data of the training user into the click prediction model to obtain the predicted click result of the training user for the training object includes:
inputting the characteristic data of the training user and the characteristic data of the training object into a GBDT model for each training user to construct a plurality of decision trees;
obtaining a predicted click result of the training user on the training object by utilizing the decision trees;
determining a combined characteristic value of the GBDT model according to the predicted click results of all decision trees;
the determining the predicted click probability of the training object according to the predicted click results of all the training users to the training object includes:
and inputting the combined characteristic value of the GBDT model into an LR model to obtain the predicted click probability of the training object.
In an alternative embodiment, the similarity between the target object and the historical object is determined according to the following formula:
Figure BDA0002563389140000031
wherein R isu,iRepresenting the rating, R, of user u on target object iu,jRepresenting the scoring of the history object j by the user u;
Figure BDA0002563389140000032
represents the average of the scores of the target object i,
Figure BDA0002563389140000033
representing the average of the scores of the historical object j. Sim (i, j) is the similarity between the target object and the history object.
In an optional embodiment, the selecting the actual associated tag of the similar object as the second candidate associated tag of the target object includes:
calculating the matching degree of each actual associated tag of the similar object with the target object;
selecting the second candidate associated tag from the actual associated tags of the similar objects according to the matching degree;
the degree of matching is calculated according to the following formula:
Figure BDA0002563389140000041
wherein S isi,NIs the similarity of the target object i and the similar object N, Ru,NAnd scoring the similar object N for the user u.
An embodiment of the present invention further provides an advertisement delivery recommendation apparatus, including:
the device comprises a determining unit, a judging unit and a judging unit, wherein the determining unit is used for determining a first candidate associated tag of a target object according to the corresponding relation between the object tag and the associated tag of the target object;
the selection unit is used for determining the similarity between the target object and the historical objects and determining the similar objects of the target object from all the historical objects according to the similarity; selecting the actual associated label of the similar object as a second candidate associated label of the target object;
the calculation unit is used for inputting the candidate associated labels and the feature data of the target object into a click prediction model aiming at each candidate associated label of the target object to obtain the predicted click probability of the candidate associated labels to the target object; the click prediction model is trained according to the characteristic data of the training sample and the actual click result to obtain corresponding model parameters;
the association unit is used for determining recommended association tags from all candidate association tags according to the predicted click probability;
and the releasing unit is used for determining the advertisement releasing strategy of the target object according to the recommendation association label.
In an optional embodiment, the selecting unit is further configured to:
determining a preliminary associated tag of the historical object according to the characteristic data of the historical object;
acquiring user behavior data in the historical object putting process;
and modifying the preliminary association tag of the historical object according to the user behavior data to obtain an actual association tag of the historical object.
In an optional embodiment, the system further includes a training unit, configured to train the click prediction model according to the following manner:
acquiring a training sample, wherein the training sample comprises characteristic data of a training object, characteristic data of a training user and an actual click result of the training user on the training object;
inputting the feature data of the training object and the feature data of the training user into the click prediction model to obtain a predicted click result of the training user for the training object;
determining the predicted click probability of the training object according to the predicted click results of all training users aiming at the training object;
determining the actual click rate of the training object according to the actual click result of the training user on the training object;
and calculating a loss function according to the actual click rate and the predicted click probability, and determining a parameter corresponding to the click prediction model when the loss function is smaller than a preset threshold value.
In an alternative embodiment, the click prediction model comprises a GBDT (gradient boosting iterative decision tree) model and an LR (logistic regression) model;
the training unit is further configured to:
inputting the characteristic data of the training user and the characteristic data of the training object into a GBDT model for each training user to construct a plurality of decision trees;
obtaining a predicted click result of the training user on the training object by utilizing the decision trees;
determining a combined characteristic value of the GBDT model according to the predicted click results of all decision trees;
and inputting the combined characteristic value of the GBDT model into an LR model to obtain the predicted click probability of the training object.
In an optional embodiment, the selecting unit is specifically configured to determine a similarity between the target object and the historical object according to the following formula:
Figure BDA0002563389140000051
wherein R isu,iRepresenting the rating, R, of user u on target object iu,jRepresenting the scoring of the history object j by the user u;
Figure BDA0002563389140000052
represents the average of the scores of the target object i,
Figure BDA0002563389140000053
representing the average of the scores of the historical object j. Sim (i, j) is the similarity between the target object and the history object.
In an optional embodiment, the selecting unit is specifically configured to:
calculating the matching degree of each actual associated tag of the similar object with the target object;
selecting the second candidate associated tag from the actual associated tags of the similar objects according to the matching degree;
the degree of matching is calculated according to the following formula:
Figure BDA0002563389140000061
wherein S isi,NIs the similarity of the target object i and the similar object N, Ru,NAnd scoring the similar object N for the user u.
An embodiment of the present invention further provides an electronic device, including:
at least one processor; and the number of the first and second groups,
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method as described above.
Embodiments of the present invention also provide a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method as described above.
In the embodiment of the invention, when the advertisement putting strategy of the target object is determined, the first candidate associated tag of the target object is determined according to the existing corresponding relation between the object tag of the target object and the associated tag. On the other hand, a certain number of history objects are selected, the similarity between the target object and each history object is determined, the similar object of the target object is selected from the history objects according to the similarity, and the second candidate associated tag of the target object is selected from the actual associated tags of the similar objects. Thus, the candidate associated tags of the target object include a first candidate associated tag and a second candidate associated tag. And inputting the candidate associated labels and the characteristic data of the target object into a click prediction model aiming at each candidate associated label of the target object to obtain the predicted click probability of the candidate associated labels to the target object. Here, the click prediction model is trained according to the feature data of the training sample (i.e., the feature data of the target object) and the actual click result to obtain corresponding model parameters. Furthermore, all the candidate associated tags can be ranked according to the predicted click probability, the recommended associated tag of the target object is determined from all the candidate associated tags, and the advertisement putting strategy of the target object is determined according to the recommended associated tag. In the embodiment of the invention, the collected characteristic data of the historical object is used as a training sample, the click prediction model is trained, the predicted click probability of the candidate associated labels of the target object is predicted, the recommended associated label is selected from all the candidate associated labels of the target object according to the predicted click probability, the advertisement putting strategy of the target object is further determined according to the recommended associated label, and the real historical object is used as the basis, so that the pertinence of advertisement putting is increased, the putting accuracy is improved, the advertisement putting of the insurance product is more accurate and appropriate, and the putting conversion rate of the insurance product can be effectively improved. Furthermore, the actual associated tag of the historical object with high similarity to the target object is used as the candidate associated tag of the target object, so that the characteristics of the target object are fully considered, and the advertisement putting strategy of the target object is analyzed and recommended from multiple aspects, thereby expanding the advertisement putting range and further improving the advertisement putting conversion rate of the insurance product.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.
FIG. 1 is a schematic diagram of a possible system architecture according to an embodiment of the present invention;
fig. 2 is a schematic flowchart of a recommendation method for advertisement delivery according to an embodiment of the present invention;
fig. 3 is a schematic diagram of a corresponding relationship between a product tag and a user tag according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a recommendation device for advertisement delivery according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail with reference to the accompanying drawings, and it is apparent that the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, a system architecture to which the embodiment of the present invention is applicable includes an advertisement delivery system 101 and a delivery channel system 102. The advertisement delivery system 101 may be a network device such as a computer, an independent device, or a server cluster formed by a plurality of servers. Preferably, the advertisement delivery system 101 may employ cloud computing technology for information processing.
The delivery channel system 102 may be a network device such as a computer, an independent device, or a server cluster formed by a plurality of servers. Preferably, the delivery channel system 102 can employ cloud computing technology for information processing.
The advertisement delivery system 101 is connected with the delivery channel system 102 through a wired or wireless network. Optionally, the wireless or wired networks described above use standard communication techniques and/or protocols. The Network is typically the Internet, but can be any Network including, but not limited to, a Local Area Network (LAN), a Metropolitan Area Network (MAN), a Wide Area Network (WAN), a mobile, wireline or wireless Network, a private Network, or any combination of virtual private networks. In some embodiments, data exchanged over the network is represented using techniques and/or formats including HyperTextMark-up Language (HTML), Extensible Markup Language (XML), and the like. All or some of the links may also be encrypted using conventional encryption techniques such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec). In other embodiments, custom and/or dedicated data communication techniques may also be used in place of, or in addition to, the data communication techniques described above.
In the method for recommending advertisement delivery provided in the embodiment of the present invention, when an advertisement delivery policy of a target object is customized, the advertisement delivery system 101 determines a first candidate associated tag of the target object according to a correspondence between an object tag of an existing target object and an associated tag. On the other hand, the advertisement delivery system 101 selects a certain number of history objects, determines the similarity between the target object and each history object, selects similar objects of the target object from the history objects according to the similarity, and selects a second candidate associated tag of the target object from the actual associated tags of the similar objects. Thus, the candidate associated tags of the target object include a first candidate associated tag and a second candidate associated tag. And inputting the candidate associated labels and the characteristic data of the target object into a click prediction model aiming at each candidate associated label of the target object to obtain the predicted click probability of the candidate associated labels to the target object. Furthermore, the advertisement delivery system 101 may rank all the candidate associated tags according to the predicted click probability, determine a recommended associated tag of the target object from all the candidate associated tags, and determine an advertisement delivery policy of the target object according to the recommended associated tag. The advertisement delivery strategy includes delivery channels, delivery time, etc. of the target objects. In this way, the advertisement delivery system 101 may send the target object to the delivery channel system 102 according to the advertisement delivery policy.
Based on the above framework, an embodiment of the present invention provides a recommendation method for advertisement delivery, as shown in fig. 2, the recommendation method for advertisement delivery provided by the embodiment of the present invention includes the following steps:
step 201, determining a first candidate associated tag of a target object according to a corresponding relationship between an object tag and an associated tag of the target object.
Step 202, determining the similarity between the target object and the historical objects, and determining the similar objects of the target object from all the historical objects according to the similarity.
Step 203, selecting the actual associated tag of the similar object as the second candidate associated tag of the target object.
Step 204: and inputting the candidate associated labels and the feature data of the target object into a click prediction model aiming at each candidate associated label of the target object to obtain the predicted click probability of the candidate associated labels to the target object.
The candidate associated labels of the target object comprise a first candidate associated label and a second candidate associated label of the target object.
And the click prediction model is trained according to the characteristic data of the training sample and the actual click result to obtain corresponding model parameters.
Specifically, feature data of a candidate associated tag and a target object are input into a click prediction model, and a predicted click result of the candidate associated tag for each feature data is obtained through calculation. And then combining the predicted click results of the candidate associated tags aiming at all the characteristic data, and calculating the predicted click probability of the candidate associated tags to the target object according to the combined predicted click results.
Step 205: and determining a recommended associated label from all the candidate associated labels according to the predicted click probability.
Step 206: and determining an advertisement putting strategy of the target object according to the recommendation association label.
In the embodiment of the invention, when the advertisement putting strategy of the target object is determined, the first candidate associated tag of the target object is determined according to the existing corresponding relation between the object tag of the target object and the associated tag. The corresponding relation between the object label and the associated label is mainly selected and set by the staff according to the characteristics of the target object and the channel characteristics. On the other hand, a certain number of history objects are selected, the similarity between the target object and each history object is determined, the similar object of the target object is selected from the history objects according to the similarity, and the second candidate associated tag of the target object is selected from the actual associated tags of the similar objects. Thus, the candidate associated tags of the target object include a first candidate associated tag and a second candidate associated tag. And inputting the candidate associated labels and the characteristic data of the target object into a click prediction model aiming at each candidate associated label of the target object to obtain the predicted click probability of the candidate associated labels to the target object. Here, the click prediction model is trained according to the feature data of the training sample and the actual click result to obtain corresponding model parameters. Furthermore, all the candidate associated tags can be ranked according to the predicted click probability, the recommended associated tag of the target object is determined from all the candidate associated tags, and the advertisement putting strategy of the target object is determined according to the recommended associated tag. In the embodiment of the invention, the collected characteristic data of the historical object is used as a training sample, the click prediction model is trained, the predicted click probability of the candidate associated labels of the target object is predicted, the recommended associated label is selected from all the candidate associated labels of the target object according to the predicted click probability, the advertisement putting strategy of the target object is further determined according to the recommended associated label, and the real historical object is used as the basis, so that the pertinence of advertisement putting is increased, the putting accuracy is improved, the advertisement putting of the insurance product is more accurate and appropriate, and the putting conversion rate of the insurance product can be effectively improved. Furthermore, the actual associated tag of the historical object with high similarity to the target object is used as the candidate associated tag of the target object, so that the characteristics of the target object are fully considered, and the advertisement putting strategy of the target object is analyzed and recommended from multiple aspects, thereby expanding the advertisement putting range and further improving the advertisement putting conversion rate of the insurance product.
The putting conversion rate is the ratio of the number of times of completing the conversion behavior to the total number of clicks of the promotion information. The calculation formula is as follows: conversion rate (number of conversions/click rate) × 100%. For example: 10 targeted users seen the targeted insurance advertisement, 5 of them clicked to jump to the target URL, and 2 of them had client funding and had subsequent conversion behavior. Then, the conversion of this generalized result is (2/5) × 100% ═ 40%.
When the insurance product is released for the first time, a preliminary release strategy is formulated, and the preliminary release strategy is to select a suitable strategy matching the product and the channel for release by business personnel according to the characteristics of the insurance product and the channel characteristics.
Specifically, the release of insurance products is mainly based on two major contents, namely, the characteristics of the insurance products themselves and the release channel characteristics. It is therefore necessary to manage and enter some information about the products and channels so that the relevant information can be traced back and, if necessary, updated.
Besides entering basic information of insurance products, such as product names, guarantee details, claim terms and processes, labeling identification needs to be carried out on the insurance products, so that transverse classification and comparison of the products can be carried out subsequently. The currently used product labels mainly have dangerous species, insurance special marks, product-oriented population (main sex, age) and the like. The target object in the embodiment of the invention is the insurance product, and the object label is the product label.
The channel management aims at carrying out relevant information entry on each delivery channel, such as channel names, channel rates, channel user characteristics, auditing key points and the like, and aiming at audience characteristics of the channel, the information needs to be determined after being deeply communicated with the channel and known. Meanwhile, the method also comprises the auditing requirement of the channel for the release so as to provide reference for the subsequent release strategy in the channel, so that the release is legal and can be effectively audited through the channel.
In a specific embodiment, the target object may be a target product, and the associated tag is a user tag.
Further, the determining an advertisement delivery policy of the target object according to the recommended associated tag includes:
and determining the delivery channel of the target product according to the recommended user label.
Specifically, the delivery channels are closely related to the user tags, and generally, a specific delivery channel is associated with a user having a specific user tag, for example, more users in the age range of 15-35 years on the a website, and if the user tag corresponding to an insurance product is "age 20-30 years", the a website can be selected as the delivery channel of the insurance product.
Therefore, in the embodiment of the invention, after the advertisement of the insurance product is put for the first time according to the initial putting strategy, the advertisement of the put insurance product is tracked, the related data is obtained, and the initial putting strategy is modified by utilizing the data to obtain the modified advertisement putting strategy, so that the advertisement putting of the insurance product is more accurate and suitable.
Further, before determining the first candidate associated tag of the target object according to the corresponding relationship between the object tag and the associated tag of the target object, the method further includes:
determining a preliminary associated tag of the historical object according to the characteristic data of the historical object;
acquiring user behavior data in the historical object putting process;
and modifying the preliminary association tag of the historical object according to the user behavior data to obtain an actual association tag of the historical object.
The preliminary association tag of the historical object in the embodiment of the invention is determined according to the characteristics of the historical object and the channel characteristics. After the historical object is released, the user behavior event of the released historical object is processed in a buried mode, the behavior data of the user aiming at the historical object is reported in real time, and then the data analysis system carries out recalculation statistics on the data every hour, so that the purpose of real-time monitoring is achieved.
Specifically, the front-end point burying technology calls a function of a point burying SDK by manually writing a code, calls an interface at a service logic function position where a point is required to be buried, and reports data of the point burying. The advertising of historical objects, i.e. insurance products, is mainly concerned with two types of indicators: data monitoring and performance monitoring.
Data monitoring, namely monitoring user information and behaviors, wherein common data monitoring items include: PV (page view), i.e. page view volume or click volume; UV (unique viewer), which refers to the number of people visiting a certain site or clicking on a different IP address of a certain news item; the user's dwell time on each page, etc.
The performance monitoring refers to monitoring the performance of the front end, and mainly includes monitoring the experience of a webpage or a product on a user end. Common performance monitoring items include: loading time of a first screen under different users, different machine types and different systems; http waits for the response time of the request; the whole downloading time of the static resources; page rendering time, etc.
According to the embodiment of the invention, the following three corresponding relations can be generated according to the user behavior data of the historical object and used as the basis for correcting the initial release strategy. The corresponding relation comprises: 1) product-user label, 2) product-channel-release time, 3) product-material AB test results.
Wherein, 1) product-user tag matching
The advertisement delivery system may retrieve all user tags and user information thereafter from the user profile repository. In the process of making an insurance product advertisement putting strategy, an insurance product and different user labels form a 1-to-N relationship. The corresponding relation can count the relation among the product, the user label and the conversion rate, and carry out real-time sequencing according to the CR (critical ratio) value of each user label, and gradually screen out the optimal user label aiming at the product maximizing the CR value.
2) Product-channel-launch time.
Because the product characteristics and the channel characteristics are different, and under the condition that the acceptance of the channel customers to the products is unknown, the transverse comparison of the matching degree of the products and the channel becomes an important reference index for subsequent delivery. Meanwhile, the advertisement putting time is an important index influencing the advertisement putting effect, the attention degree can be greatly reduced by wrong putting time, and the conversion rate is reduced. The corresponding relation is used for counting the releasing CR values of the products in different channels and different releasing time and sequencing. And forming a matching and sequencing relation of product-product label-channel-delivery time as a basis for subsequently modifying the advertisement delivery strategy.
3) product-Material AB test
Because advertising conversion is a process that attracts the attention of customers in a short time and converts customers successfully, materials (i.e., advertising pages, dialects, etc.) are important faces and tools for promoting conversion. Generally, an AB test is used to determine which advertisement interface is more attractive and can effectively transform the client.
That is to say, in the embodiment of the present invention, data collection is performed on delivered history objects, so as to generate a "product-user label" correspondence and a "product-channel" correspondence of the history objects. Therefore, the advertisement putting strategy of the historical object is modified, the preliminary associated label in the historical object is modified, and the actual associated label is obtained.
Further, in the embodiment of the present invention, the candidate associated tag of the target object further includes a second candidate associated tag in addition to the first candidate associated tag preliminarily selected according to the characteristics of the target object and the channel characteristics. The second candidate associated label is determined from the actual associated labels of the historical objects with higher similarity to the target object.
Wherein the similarity between the target object and the historical object is determined according to the following formula:
Figure BDA0002563389140000141
wherein R isu,iRepresenting the rating, R, of user u on target object iu,jRepresenting the scoring of the history object j by the user u.
Figure BDA0002563389140000142
Represents the average of the scores of the target object i,
Figure BDA0002563389140000143
representing the average of the scores of the historical object j. Sim (i, j) is the similarity between the target object and the history object.
The average value of the user scores, wherein some users are biased to give a high score and some users are biased to give a low score, eliminates the influence of the scoring habits of different users by subtracting the average value of the user scores. The closer the value of Sim (i, j) is to 1, the higher the similarity between the target object and the history object.
And according to the similarity, after the similar object of the target object is determined, determining a second candidate associated tag from the actual associated tags of the similar object.
Preferably, the selecting the actual associated tag of the similar object as the second candidate associated tag of the target object includes:
calculating the matching degree of each actual associated tag of the similar object with the target object;
selecting the second candidate associated tag from the actual associated tags of the similar objects according to the matching degree;
the degree of matching is calculated according to the following formula:
Figure BDA0002563389140000144
wherein S isi,NIs the similarity of the target object i and the similar object N, Ru,NAnd scoring the similar object N for the user u.
And carrying out weighted summation on the scores of the objects scored by the user u, wherein the weight is the similarity between each similar object and the target object, then averaging the sum of all the similarities, and calculating to obtain the score of the user u on the target object i. Wherein, the higher the score, the higher the matching degree of the user to the target object i (candidate associated tag).
Specifically, all the actual associated tags of the similar objects may be used as the second candidate associated tags of the target object, or a part of the actual associated tags of the similar objects may be selected as the second candidate associated tags of the target object, where the selection is performed in a manner of calculating the matching degree between each actual associated tag of the similar objects and the target object, so as to select the corresponding actual associated tag as the second candidate associated tag according to the matching degree. In addition, the number of similar objects corresponding to the target object is not limited in the embodiment of the present invention, that is, one similar object may be used, or multiple similar objects may be used.
For example, as shown in fig. 3, product C is the target object, and the history objects are product a and product B. And calculating the similarity between the product C and the product A and the similarity between the product C and the product B by using the formula 1, determining that the similarity between the product A and the product C meets the condition according to the calculation result, and taking the product A as a similar object of the product C. Since the product labels of the product a are 1, 2 and 3, and the corresponding user labels are a, b, C, d, e and g, respectively, the user labels a, b, C, d, e and g are taken as second candidate associated labels of the product C. In addition, since the first candidate associated tag of the product C is the user tag b, C, e, g, the candidate associated tag of the product C is a, b, C, d, e, g.
After the candidate associated labels of the target object are obtained, the predicted click probability corresponding to each candidate associated label is output by using the click prediction model, namely, the candidate associated labels of the target object are input into the click prediction model, and the predicted click probability of each candidate associated label is calculated through the click prediction model. And therefore, the recommended associated label can be selected from all the candidate associated labels according to the predicted click probability. And establishing an incidence relation between the target object and the recommended incidence label, namely establishing an incidence relation between the target product and the user label, and determining a channel for releasing the target product according to the user label.
Preferably, in the embodiment of the present invention, the click prediction model is obtained by training according to the following manner:
acquiring a training sample, wherein the training sample comprises characteristic data of a training object, characteristic data of a training user and an actual click result of the training user on the training object;
inputting the feature data of the training object and the feature data of the training user into the click prediction model to obtain a predicted click result of the training user for the training object;
determining the predicted click probability of the training object according to the predicted click results of all training users aiming at the training object;
determining the actual click rate of the training object according to the actual click result of the training user on the training object;
and calculating a loss function according to the actual click rate and the predicted click probability, and determining a parameter corresponding to the click prediction model when the loss function is smaller than a preset threshold value.
The click prediction model includes a GBDT (Gradient Boosting Decision Tree) model and an LR (Logistic regression) model.
The inputting the feature data of the training object and the feature data of the training user into the click prediction model to obtain the predicted click result of the training user for the training object includes:
inputting the characteristic data of the training user and the characteristic data of the training object into a GBDT model for each training user to construct a plurality of decision trees;
obtaining a predicted click result of the training user on the training object by utilizing the decision trees;
determining a combined characteristic value of the GBDT model according to the predicted click results of all decision trees;
the determining the predicted click probability of the training object according to the predicted click results of all the training users to the training object includes:
and inputting the combined characteristic value of the GBDT model into an LR model to obtain the predicted click probability of the training object.
Since the GBDT model alone is easy to be over-fitted, the GBDT model and the LR model are combined for training in the embodiment of the present invention, and the training process of the GBDT model and the LR model is described in detail below.
GBDT is an iterative decision tree algorithm, consisting of several decision trees. The gradient lifting tree model is a model combining a decision tree and a lifting method (such as XGboost), and the core idea is as follows: in the process of constructing a series of decision trees, the subsequent decision tree learns the conclusions and residuals of all the previous decision trees, when the residuals of the subsequent decision tree are smaller than a set threshold value or reach the iteration times, the model terminates training, and a plurality of decision trees are finally obtained by continuously fitting the residuals of the previous decision trees. And for the sample to be detected, the output result of the gradient lifting tree model is the sum of the output results of the k decision trees. In this process, the samples are randomly selected and the features are randomly selected, which means that some samples in the total training set may appear in the training set of one tree more times or never appear in the training set of any one tree. The method mainly comprises the following steps: and (3) randomly extracting n sample sets from m model training samples by applying a resampling technology, and constructing n decision trees. In the growth process of each decision tree, each node randomly extracts F features from all the features as a subset of current node splitting, and the minimum mean square error is usually adopted as a splitting judgment standard when the decision tree is constructed, so that the best splitting mode is selected. And combining the n decision trees into a final GBDT model.
In the embodiment of the invention, the corresponding user can be determined aiming at the target product, and then the advertisement is pushed to the determined user; the corresponding target product may also be determined for a particular user. Here, taking the example of determining the corresponding user for the target product, the training process of the click probability model is introduced.
First, training samples are obtained. The training sample form in the embodiment of the invention is { (X)1，Y1)，(X2，Y2)，...(Xn，Yn]In which XnMultidimensional feature data for the nth training subject, e.g. X1Multi-dimensional feature data for the 1 st training object, X2The 2 nd training object. The multi-dimensional feature data is in the form of (x)n1，xn2，xn3，...xni) Wherein i is the ith feature of the nth training object. For example, xn1Representing the age, x, of the user corresponding to the training subjectn2May represent the gender of the user corresponding to the training subject. Y isnIndicating whether the user recommends the nth training object or not, if yes, YnIs 1; if not, YnIs 0.
Inputting the training samples into a GBDT model to generate n decision trees to form a regression tree f (x).
And in the training process, inputting the training samples into the GBDT model. Specifically, in an input space where a training data set is located, each region is recursively divided into two sub-regions, an output value on each sub-region is determined, and a two-branch decision tree is constructed;
(1) selecting an optimal segmentation variable j and an optimal segmentation point s, and solving the following formula:
Figure BDA0002563389140000181
(2) the variable j is traversed, the fixed segmentation variable j is scanned for the segmentation point s, and the value (j, s) that minimizes equation 3 is selected.
Dividing the region by the selected (j, s) and determining the corresponding output value according to the following formula:
R1(j，s)＝{x|x(j)≤s}，R2(j，s)＝{x|x(j)≤s}
Figure BDA0002563389140000182
(3) continuing to call the steps (1) and (2) for the two sub-areas until a stopping condition is met;
(4) dividing an input space into M regions R1，R2，...，RMGenerating a 1 st decision tree:
Figure BDA0002563389140000183
(5) a 2 nd decision tree is generated. Constructing training sample data of the 2 nd decision Tree, representing target variables of each training sample by using residual errors, And specifically, generating the 2 nd decision Tree by using a CART (Classification And Regression Tree) algorithm. The 2 nd decision tree is defined as DT2，DT2(Xi) Represents the 2 nd decision tree pair training sample XiThe training samples input into the 2 nd decision tree are as follows:
{(X1，Y1-DT1(X1))，(X2，Y2-DT1(X2))，...(Xn，Yn-DT1(Xn) Equation 6
Initializing the weak learner:
Figure BDA0002563389140000184
for the number of iteration rounds M ═ 1, 2, ·, M there are:
for each training sample i 1, 2, ·, N, a negative gradient, i.e. a residual, is calculated according to the following formula:
Figure BDA0002563389140000191
using the obtained residual error as a new true value of the sample, and using the data (x)i，rim) (i-1, 2.. N) as training data of the next decision tree, a new regression tree f is obtainedm(x) In that respect The corresponding leaf node region is RjmJ is 1, 2. Wherein J is the number of leaf nodes of the regression tree t.
The best fit value is calculated for leaf region J1, 2.. J according to the following formula:
Figure BDA0002563389140000192
the strong learner is updated according to the following formula:
Figure BDA0002563389140000193
the strong learner is obtained according to the following formula:
Figure BDA0002563389140000194
(6) and in the same way, generating the m decision tree. The training sample data for constructing the mth decision tree is as follows:
{(X1，DTm-2(X1)-DTm-1(X1))，(X2，DTm-2(X2)-DTm-1(X2))，...，(Xn，DTm-2(X1)-DTm-1(Xn))}
thus, m decision trees are generated by using the CART algorithm, wherein the mth decision tree is defined as DTm，DTm(Xi) Represents the mth decision tree pair training sample XiThe predicted click result of (1). The predicted result of each decision tree is added up to be the predicted value of the GBDT model, which is in the form of (0110111.). Wherein, 1 represents that the predicted click result is recommended, and 0 represents not recommended.
And (4) taking leaf nodes of the training samples in the GBDT model as output, inputting the output into the LR model, and obtaining the predicted click probability of each user label on the training samples.
And finally, calculating a loss function according to the actual click rate of the user label on the training sample and the predicted click probability, and determining a parameter corresponding to the click prediction model when the loss function is smaller than a preset threshold value, such as a dividing point of each decision tree in the regression tree. According to the embodiment of the invention, the parameters corresponding to the click prediction model are changed according to the change of the training samples, and with the increase of the number of the training samples in the input click prediction model, the parameters of the click prediction model are more suitable, and the prediction result of the click prediction model is more accurate.
Still taking fig. 3 as an example, the candidate user tags a, b, C, d, e, and g of the product C are input into the click prediction model, and are ranked as C, b, e, d, a, and g according to the obtained predicted click probability. Therefore, the corresponding channel can be selected for advertisement putting according to the sequence.
In addition, the embodiment of the invention can also carry out advertisement putting from the user dimension, namely, the corresponding target product is determined aiming at the specific user. The user dimension is similar to the product dimension principle, the basic idea being that if in historical releases user a matches product label a, user B matches product labels a, B, C, and user C matches product labels a and C, then user a is considered similar to users B and C because they both match a. The user A, B, C will have a common user label for a in the engine, such as "love for outdoor sports". And the users matching a also match c, so c is recommended to the user A, and the c is the 'sports accident risk' of the product label. Because the user labels are very large, the calculation of the complete set is time-consuming and labor-consuming, and is not efficient enough. Thus, rather than computing globally directly, user tags are first filtered according to intent and goal. Moreover, product labels can be synchronously limited, and the range of the content library is further reduced. Content is cut off according to the delivery requirement or the insurant group characteristics (input during product management) oriented by insurance products, a small part of content meeting the target is efficiently screened out from a large content library, a massive content library which cannot be grasped is changed into a relatively small content library which can be grasped, and then the content enters a recommendation model. This effectively balances computational cost and effectiveness. For example, if the release is mainly targeted at a 30-40 year old female user with children going to operate and expand, the user tags that can be determined would be: the system comprises a user identification module, a user. If further needed, secondary selection can be carried out on the product label, such as 'female health risk', 'children accident risk' and the like, and the recommendation result is calculated more finely. The final conclusion may be that customers of such tags are more willing to invest in children than to focus on their health (the engine calculates that the product tags for children's insurance class score is higher and ranked further up under such tag users). Such users are more active in a channel and are more likely to come online in the channel after 20:00 a night, see the promotional advertisements and make conversions.
An embodiment of the present invention further provides a recommendation apparatus for advertisement delivery, as shown in fig. 4, including:
a determining unit 401, configured to determine a first candidate associated tag of a target object according to a corresponding relationship between an object tag and an associated tag of the target object;
a selecting unit 402, configured to determine similarity between a target object and history objects, and determine similar objects of the target object from all history objects according to the similarity; selecting the actual associated label of the similar object as a second candidate associated label of the target object;
a calculating unit 403, configured to, for each candidate associated tag of the target object, input the candidate associated tag and feature data of the target object into a click prediction model, so as to obtain a predicted click probability of the candidate associated tag on the target object; the click prediction model is trained according to the characteristic data of the training sample and the actual click result to obtain corresponding model parameters;
an association unit 404, configured to determine a recommended association tag from all candidate association tags according to the predicted click probability;
and an advertisement putting unit 405, configured to determine an advertisement putting policy of the target object according to the recommended associated tag.
In an alternative embodiment, the selecting unit 402 is further configured to:
determining a preliminary associated tag of the historical object according to the characteristic data of the historical object;
acquiring user behavior data in the historical object putting process;
and modifying the preliminary association tag of the historical object according to the user behavior data to obtain an actual association tag of the historical object.
In an alternative embodiment, the system further includes a training unit 406, configured to train the click prediction model according to the following manner:
acquiring a training sample, wherein the training sample comprises characteristic data of a training object, characteristic data of a training user and an actual click result of the training user on the training object;
inputting the feature data of the training object and the feature data of the training user into the click prediction model to obtain a predicted click result of the training user for the training object;
determining the predicted click probability of the training object according to the predicted click results of all training users aiming at the training object;
determining the actual click rate of the training object according to the actual click result of the training user on the training object;
and calculating a loss function according to the actual click rate and the predicted click probability, and determining a parameter corresponding to the click prediction model when the loss function is smaller than a preset threshold value.
In an alternative embodiment, the click prediction model comprises a GBDT (gradient boosting iterative decision tree) model and an LR (logistic regression) model;
the training unit 406 is further configured to:
inputting the characteristic data of the training user and the characteristic data of the training object into a GBDT model for each training user to construct a plurality of decision trees;
obtaining a predicted click result of the training user on the training object by utilizing the decision trees;
determining a combined characteristic value of the GBDT model according to the predicted click results of all decision trees;
and inputting the combined characteristic value of the GBDT model into an LR model to obtain the predicted click probability of the training object.
In an optional embodiment, the selecting unit 402 is specifically configured to determine the similarity between the target object and the historical object according to the following formula:
Figure BDA0002563389140000221
wherein R isu,iRepresenting the rating, R, of user u on target object iu,jRepresenting the scoring of the history object j by the user u.
Figure BDA0002563389140000222
Represents the average of the scores of the target object i,
Figure BDA0002563389140000223
representing the average of the scores of the historical object j. Sim (i, j) is the similarity between the target object and the history object.
In an alternative embodiment, the selecting unit 402 is specifically configured to:
calculating the matching degree of each actual associated tag of the similar object with the target object;
selecting the second candidate associated tag from the actual associated tags of the similar objects according to the matching degree;
the degree of matching is calculated according to the following formula:
Figure BDA0002563389140000231
wherein S isi,NIs the similarity of the target object i and the similar object N, Ru,NAnd scoring the similar object N for the user u.
Based on the same principle, the present invention also provides an electronic device, as shown in fig. 5, including:
the system comprises a processor 501, a memory 502, a transceiver 503 and a bus interface 504, wherein the processor 501, the memory 502 and the transceiver 503 are connected through the bus interface 504;
the processor 501 is configured to read the program in the memory 502, and execute the following method:
determining a first candidate associated tag of a target object according to the corresponding relation between the object tag and the associated tag of the target object;
determining the similarity between a target object and historical objects, and determining the similar objects of the target object from all the historical objects according to the similarity;
selecting the actual associated label of the similar object as a second candidate associated label of the target object;
inputting the candidate associated labels and the feature data of the target object into a click prediction model aiming at each candidate associated label of the target object to obtain the predicted click probability of the candidate associated labels to the target object; the click prediction model is trained according to the characteristic data of the training sample and the actual click result to obtain corresponding model parameters;
determining recommended associated labels from all candidate associated labels according to the predicted click probability;
and determining an advertisement putting strategy of the target object according to the recommendation association label.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims (9)

1. An advertisement placement recommendation method, comprising:
determining a first candidate associated tag of a target object according to the corresponding relation between the object tag and the associated tag of the target object;
determining the similarity between a target object and historical objects, and determining the similar objects of the target object from all the historical objects according to the similarity;
selecting the actual associated label of the similar object as a second candidate associated label of the target object;
inputting the candidate associated labels and the feature data of the target object into a click prediction model aiming at each candidate associated label of the target object to obtain the predicted click probability of the candidate associated labels to the target object; the click prediction model is trained according to the characteristic data of the training sample and the actual click result to obtain corresponding model parameters;
determining recommended associated labels from all candidate associated labels according to the predicted click probability;
and determining an advertisement putting strategy of the target object according to the recommendation association label.
2. The method of claim 1, wherein before determining the first candidate associated tag of the target object according to the correspondence between the object tag and the associated tag of the target object, the method further comprises:
determining a preliminary associated tag of the historical object according to the characteristic data of the historical object;
acquiring user behavior data in the historical object putting process,
and modifying the preliminary association tag of the historical object according to the user behavior data to obtain an actual association tag of the historical object.
3. The method of claim 1, wherein the click prediction model is trained according to:
acquiring a training sample, wherein the training sample comprises characteristic data of a training object, characteristic data of a training user and an actual click result of the training user on the training object;
inputting the feature data of the training object and the feature data of the training user into the click prediction model to obtain a predicted click result of the training user for the training object;
determining the predicted click probability of the training object according to the predicted click results of all training users aiming at the training object;
determining the actual click rate of the training object according to the actual click result of the training user on the training object;
and calculating a loss function according to the actual click rate and the predicted click probability, and determining a parameter corresponding to the click prediction model when the loss function is smaller than a preset threshold value.
4. The method of claim 3, in which the click prediction model comprises a GBDT gradient boosting iterative decision tree model and a LR logistic regression model;
the inputting the feature data of the training object and the feature data of the training user into the click prediction model to obtain the predicted click result of the training user for the training object includes:
inputting the characteristic data of the training user and the characteristic data of the training object into a GBDT model for each training user to construct a plurality of decision trees;
obtaining a predicted click result of the training user on the training object by utilizing the decision trees;
determining a combined characteristic value of the GBDT model according to the predicted click results of all decision trees;
the determining the predicted click probability of the training object according to the predicted click results of all the training users to the training object includes:
and inputting the combined characteristic value of the GBDT model into an LR model to obtain the predicted click probability of the training object.
5. The method of any of claims 1-4, wherein the similarity of the target object to the historical object is determined according to the following formula:
Figure FDA0002563389130000021
wherein R isu,iRepresenting the rating, R, of user u on target object iu,jRepresenting the scoring of the history object j by the user u;
Figure FDA0002563389130000022
represents the average of the scores of the target object i,
Figure FDA0002563389130000023
represents the average of the scores of the historical object j; sim (i, j) is the similarity between the target object and the history object.
6. The method of claim 1, wherein said selecting the actual associated tag of the similar object as the second candidate associated tag of the target object comprises:
calculating the matching degree of each actual associated tag of the similar object with the target object;
selecting the second candidate associated tag from the actual associated tags of the similar objects according to the matching degree;
the degree of matching is calculated according to the following formula:
Figure FDA0002563389130000031
wherein S isi,NIs the similarity of the target object i and the similar object N, Ru,NAnd scoring the similar object N for the user u.
7. An advertisement placement recommendation device, comprising:
the device comprises a determining unit, a judging unit and a judging unit, wherein the determining unit is used for determining a first candidate associated tag of a target object according to the corresponding relation between the object tag and the associated tag of the target object;
the selection unit is used for determining the similarity between the target object and the historical objects and determining the similar objects of the target object from all the historical objects according to the similarity; selecting the actual associated label of the similar object as a second candidate associated label of the target object;
the calculation unit is used for inputting the candidate associated labels and the feature data of the target object into a click prediction model aiming at each candidate associated label of the target object to obtain the predicted click probability of the candidate associated labels to the target object; the click prediction model is trained according to the characteristic data of the training sample and the actual click result to obtain corresponding model parameters;
the association unit is used for determining recommended association tags from all candidate association tags according to the predicted click probability;
and the releasing unit is used for determining the advertisement releasing strategy of the target object according to the recommendation association label.
8. An electronic device, comprising:
at least one processor; and the number of the first and second groups,
a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,
the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.
9. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the method of any one of claims 1 to 6.
CN202010622122.XA 2020-06-30 2020-06-30 Advertisement delivery recommendation method and device Pending CN111768244A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010622122.XA CN111768244A (en) 2020-06-30 2020-06-30 Advertisement delivery recommendation method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010622122.XA CN111768244A (en) 2020-06-30 2020-06-30 Advertisement delivery recommendation method and device

Publications (1)

Publication Number Publication Date
CN111768244A true CN111768244A (en) 2020-10-13

Family

ID=72723283

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010622122.XA Pending CN111768244A (en) 2020-06-30 2020-06-30 Advertisement delivery recommendation method and device

Country Status (1)

Country Link
CN (1) CN111768244A (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112270578A (en) * 2020-11-23 2021-01-26 支付宝(杭州)信息技术有限公司 Object display method and device and electronic equipment
CN112508615A (en) * 2020-12-10 2021-03-16 深圳市欢太科技有限公司 Feature extraction method, feature extraction device, storage medium, and electronic apparatus
CN112508284A (en) * 2020-12-10 2021-03-16 网易（杭州）网络有限公司 Display material preprocessing method, putting method, system, device and equipment
CN112767053A (en) * 2021-01-29 2021-05-07 北京达佳互联信息技术有限公司 Information processing method, information processing device, electronic equipment and storage medium
CN112989023A (en) * 2021-03-25 2021-06-18 北京百度网讯科技有限公司 Label recommendation method, device, equipment, storage medium and computer program product
CN113011911A (en) * 2021-01-21 2021-06-22 腾讯科技（深圳）有限公司 Data prediction method, device, medium and electronic equipment based on artificial intelligence
CN113763107A (en) * 2021-01-26 2021-12-07 北京沃东天骏信息技术有限公司 Object information pushing method, device, equipment and storage medium

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112270578A (en) * 2020-11-23 2021-01-26 支付宝(杭州)信息技术有限公司 Object display method and device and electronic equipment
CN112508615A (en) * 2020-12-10 2021-03-16 深圳市欢太科技有限公司 Feature extraction method, feature extraction device, storage medium, and electronic apparatus
CN112508284A (en) * 2020-12-10 2021-03-16 网易（杭州）网络有限公司 Display material preprocessing method, putting method, system, device and equipment
CN113011911A (en) * 2021-01-21 2021-06-22 腾讯科技（深圳）有限公司 Data prediction method, device, medium and electronic equipment based on artificial intelligence
CN113763107A (en) * 2021-01-26 2021-12-07 北京沃东天骏信息技术有限公司 Object information pushing method, device, equipment and storage medium
CN112767053A (en) * 2021-01-29 2021-05-07 北京达佳互联信息技术有限公司 Information processing method, information processing device, electronic equipment and storage medium
CN112989023A (en) * 2021-03-25 2021-06-18 北京百度网讯科技有限公司 Label recommendation method, device, equipment, storage medium and computer program product
CN112989023B (en) * 2021-03-25 2023-07-28 北京百度网讯科技有限公司 Label recommendation method, device, equipment, storage medium and computer program product

Similar Documents

Publication Publication Date Title
CN111768244A (en) Advertisement delivery recommendation method and device
US11004107B2 (en) Target user directing method and apparatus and computer storage medium
US9245252B2 (en) Method and system for determining on-line influence in social media
CN108427708B (en) Data processing method, data processing apparatus, storage medium, and electronic apparatus
CN106997549A (en) The method for pushing and system of a kind of advertising message
CN108777701B (en) Method and device for determining information audience
EP3189449A2 (en) Sentiment rating system and method
CN105095444A (en) Information acquisition method and device
US20110196690A1 (en) Method and system of selecting landing pages and optimizing routing efficiency
CN112148973B (en) Data processing method and device for information push
CN110175857B (en) Method and device for determining optimal service
EP3783522A1 (en) Semantic model instantiation method, system and device
CN112270569A (en) Advertisement putting method and device, electronic equipment and storage medium
US20210342744A1 (en) Recommendation method and system and method and system for improving a machine learning system
CN114861050A (en) Feature fusion recommendation method and system based on neural network
CN108595580B (en) News recommendation method, device, server and storage medium
CN115222433A (en) Information recommendation method and device and storage medium
CN112989053A (en) Periodical recommendation method and device
WO2023082864A1 (en) Training method and apparatus for content recommendation model, device, and storage medium
CN114399352A (en) Information recommendation method and device, electronic equipment and storage medium
CN115293867A (en) Financial reimbursement user portrait optimization method, device, equipment and storage medium
CN111400413B (en) Method and system for determining category of knowledge points in knowledge base
CN111400567B (en) AI-based user data processing method, device and system
CN115114073A (en) Alarm information processing method and device, storage medium and electronic equipment
CN112818215A (en) Product data processing method, device, equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination