CN113743636A - Target operation prediction method and device, electronic equipment and storage medium - Google Patents
Target operation prediction method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN113743636A CN113743636A CN202010469618.8A CN202010469618A CN113743636A CN 113743636 A CN113743636 A CN 113743636A CN 202010469618 A CN202010469618 A CN 202010469618A CN 113743636 A CN113743636 A CN 113743636A
- Authority
- CN
- China
- Prior art keywords
- account
- target operation
- historical
- data
- characteristic data
- 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
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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Abstract
The disclosure provides a target operation prediction method, a target operation prediction device, electronic equipment and a storage medium, relates to the technical field of computers, and is used for optimizing a process of predicting target operation. The method comprises the following steps: acquiring operation intention characteristic data of an account to be predicted, wherein the operation intention characteristic data comprises characteristic data representing the preference degree of the account on target operation, and the target operation comprises operation performed after the account browses a push object; estimating a second degree of association between the operation intention characteristics of the account to be predicted and a target operation result according to the first degree of association between the operation intention characteristics of each historical account and the target operation result determined by the duration data of the pushing object browsed by each historical account; and converting the estimated second relevance into an operation intention probability value of the account to be predicted for performing the target operation, wherein the target operation result comprises a result of whether the target operation is performed or not. The method provides a new way of predicting the target operation.
Description
Technical Field
The present disclosure relates to the field of computer technologies, and in particular, to a target operation prediction method and apparatus, an electronic device, and a storage medium.
Background
In the related art, when it is predicted whether a user performs a specified operation after aiming at a push object, for example, when the user is predicted to purchase a push commodity in an internet e-commerce, historical data including user operation characteristic data and an operation result of whether the user performs the specified operation is obtained, and a data association relation between the user operation characteristic data and the operation result of whether the user performs the specified operation is analyzed based on the obtained historical data, so that whether the account to be predicted performs the specified operation aiming at the push object is estimated based on the data association relation and the user operation characteristic data of the user to be predicted; if the pushed object is a commodity and the appointed operation is commodity purchasing, historical data containing commodity purchasing results need to be acquired, and whether the user purchases the pushed commodity is predicted based on the commodity purchasing results and user operation characteristic data in the acquired historical data; if the pushing object is an application and the designated operation is downloading the application, historical data containing the application downloading result needs to be acquired, and then whether the user can download the pushed application is estimated based on application downloading information and user operation characteristic data in the acquired historical data.
However, in the above prediction process, it is necessary to obtain history data including operation results of whether the user performs the specified operation, and these history data are usually stored in a server of a specific platform, for example, history data including commodity purchase results are stored in a server of an e-commerce platform, history data including application download results are stored in a server of an application operation platform, and authorization of the specific platform is required to obtain these history data.
Disclosure of Invention
The embodiment of the disclosure provides a target operation prediction method, a target operation prediction device, an electronic device and a storage medium, which are used for providing a new method for predicting a target operation.
In a first aspect of the present disclosure, a target operation prediction method is provided, including:
acquiring operation intention characteristic data of an account to be predicted, wherein the operation intention characteristic data comprises characteristic data representing the preference degree of the account on target operation, and the target operation comprises operation performed after the account browses a push object;
estimating a second degree of association between the operation intention characteristic data of the account to be predicted and a target operation result according to the first degree of association of the operation intention characteristic data of each historical account and the target operation result determined by the duration data of the pushing object browsed by each historical account; and converting the estimated second relevance into an operation intention probability value of the account to be predicted for performing the target operation, wherein the target operation result comprises a result of whether the target operation is performed or not.
In a possible implementation manner, the second association degree between the operation intention characteristic data of the account to be predicted and the target operation result is estimated according to the first association degree of the target operation result determined by the operation intention characteristic data of each historical account and the time length data of the pushing object browsed by each historical account; and converting the estimated second relevance into an operation intention probability value of the account to be predicted for performing the target operation, wherein the step comprises the following steps of:
inputting the operation intention characteristic data of the account to be predicted by adopting a trained operation intention prediction model, and obtaining the operation intention probability value of the account to be predicted for carrying out the target operation, which is output by the operation intention prediction model, wherein the operation intention prediction model is obtained by training by adopting the operation intention characteristic data of each historical account and the time length data of the object to be pushed browsed and pushed by each historical account as training samples based on a machine learning method.
In one possible implementation, the trained operation intention prediction model is obtained by:
extracting operation intention characteristic data of the historical account and duration data of the historical account browsing and pushing objects in each training sample;
according to the duration data of the pushed object browsed by the historical account in each training sample, determining a target operation result corresponding to each training sample;
and obtaining the trained operation intention estimation model by utilizing the mapping relation between the operation intention characteristic data of the historical account in each training sample and the corresponding target operation result and adjusting the operation intention estimation model obtained by historical training or the initially created operation intention estimation model through machine learning.
In a possible implementation manner, the step of determining a target operation result of the historical object in each training sample according to the duration data of the pushed object browsed by the historical account in each training sample includes:
according to the time length data of the historical account browsing the pushed object in each training sample, determining the target operation result corresponding to the training sample with the time length of the historical account browsing the pushed object not less than the set time length threshold value as the first result of the target operation, and
and determining a target operation result corresponding to the training sample with the duration of the historical account browsing the pushed object being less than the set duration threshold as a second result of not performing the target operation.
In one possible implementation, the method further includes: and determining a target operation result corresponding to a training sample without the duration data of the historical account browsing push object as a second result without the target operation.
In a possible implementation manner, the set duration threshold is determined according to duration data of a first set number of historical objects browsing the pushed object and a corresponding real target operation result.
In one possible implementation, the method further includes:
determining historical browsing object data within a set time length before the current time as the training sample; or
And determining a second set amount of historical browsing object data before the current time as the training sample.
In a second aspect of the present disclosure, a target operation prediction apparatus is provided, including:
the intention characteristic obtaining unit is configured to execute obtaining operation intention characteristic data of the account to be predicted, the operation intention characteristic data comprise characteristic data representing the preference degree of the account on target operation, and the target operation comprises operation performed after the account browses a push object;
the target operation prediction unit is configured to execute a first association degree of a target operation result determined by operation intention characteristic data of each historical account and duration data of browsing a push object by each historical account, and estimate a second association degree of the operation intention characteristic data of the account to be predicted and the target operation result; and converting the estimated second relevance into an operation intention probability value of the account to be predicted for performing the target operation, wherein the target operation result comprises a result of whether the target operation is performed or not.
In a possible implementation, the target operation prediction unit is specifically configured to perform:
inputting the operation intention characteristic data of the account to be predicted by adopting a trained operation intention prediction model, and obtaining the operation intention probability value of the account to be predicted for carrying out the target operation, which is output by the operation intention prediction model, wherein the operation intention prediction model is obtained by training by adopting the operation intention characteristic data of each historical account and the time length data of the object to be pushed browsed and pushed by each historical account as training samples based on a machine learning method.
In one possible implementation, the target operation prediction unit is further configured to perform:
obtaining the trained operation intention prediction model by the following method:
extracting operation intention characteristic data of the historical account and duration data of the historical account browsing and pushing objects in each training sample;
according to the duration data of the pushed object browsed by the historical account in each training sample, determining a target operation result corresponding to each training sample;
and obtaining the trained operation intention estimation model by utilizing the mapping relation between the operation intention characteristic data of the historical account in each training sample and the corresponding target operation result and adjusting the operation intention estimation model obtained by historical training or the initially created operation intention estimation model through machine learning.
In a possible implementation, the target operation prediction unit is specifically configured to perform:
according to the time length data of the historical account browsing the pushed object in each training sample, determining the target operation result corresponding to the training sample with the time length of the historical account browsing the pushed object not less than the set time length threshold value as the first result of the target operation, and
and determining a target operation result corresponding to the training sample with the duration of the historical account browsing the pushed object being less than the set duration threshold as a second result of not performing the target operation.
In one possible implementation, the target operation prediction unit is further configured to perform:
and determining a target operation result corresponding to a training sample without the duration data of the historical account browsing push object as a second result without the target operation.
In a possible implementation manner, the set duration threshold is determined according to duration data of a first set number of historical objects browsing the pushed object and a corresponding real target operation result.
In one possible implementation, the target operation prediction unit is further configured to perform:
determining historical browsing object data within a set time length before the current time as the training sample; or
And determining a second set amount of historical browsing object data before the current time as the training sample.
In a third aspect of the present disclosure, a computer device is provided, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the method of any one of the first aspect and one possible implementation manner is implemented.
In a fourth aspect of the present disclosure, a computer-readable storage medium is provided, which stores computer instructions that, when executed on a computer, cause the computer to perform the method according to any one of the first aspect and one of the possible embodiments.
The scheme of the present disclosure brings at least the following beneficial effects:
according to the method, the label of the operation result of the target operation of the training sample is determined according to the duration of the browsing object, the second association degree of the operation intention characteristic data of the account to be predicted and the target operation result is estimated by utilizing the operation intention characteristic data of the historical account in the training sample and the first association degree of the target operation result determined by the duration data of the pushing object browsed by the historical account, the second association degree is converted into the operation intention probability value of the target operation after the pushing object browsed by the account to be predicted, the data of the real target operation result of whether the historical account performs the target operation or not does not need to be acquired from a specific platform in the process, and a new target operation predicting mode is provided.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure and are not to be construed as limiting the disclosure.
Fig. 1 is a schematic diagram of an application scenario provided in an exemplary embodiment of the present disclosure;
FIG. 2 is a flowchart illustrating a method for predicting a target operation according to an exemplary embodiment of the disclosure;
fig. 3 is a schematic diagram illustrating an acquisition process of a currently used machine learning model according to an exemplary embodiment of the present disclosure;
fig. 4 is a schematic diagram illustrating another currently used acquisition process of a machine learning model according to an exemplary embodiment of the present disclosure;
fig. 5 is a schematic flowchart of a process for obtaining a set duration threshold according to an exemplary embodiment of the present disclosure;
fig. 6 is a schematic structural diagram of a target operation prediction apparatus according to an exemplary embodiment of the present disclosure;
fig. 7 is a schematic structural diagram of an electronic device according to an exemplary embodiment of the present disclosure.
Detailed Description
In order to make the technical solutions of the present disclosure better understood by those of ordinary skill in the art, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.
In order to facilitate better understanding of the technical solutions of the present disclosure by those skilled in the art, the following technical terms related to the present disclosure are explained.
The terms "first," "second," and the like in the description and in the claims of the present disclosure and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the disclosure described herein are capable of operation in sequences other than those illustrated or otherwise described herein.
Pushing an object: including objects pushed to accounts over a network. The push object in the embodiment of the present disclosure may include, but is not limited to, a website link, that is, the push object may be an internet e-commerce level advertisement object detail page, such as a commodity detail page corresponding to a commodity link or an application detail page corresponding to an application link, and the application may be, but is not limited to, one or more of a game application, an information application, and an instant messaging application; the account includes an identification of the user on a specific platform, and in the embodiment of the disclosure, the account is sometimes expressed synonymously with the user.
Target operation: the method comprises the operation performed after the account browses the push object. If the pushing object is a commodity detail page, the target operation may be an operation of adding a commodity corresponding to the commodity detail page into an electronic shopping cart, or the target operation may also be an operation of transferring electronic resources for the commodity corresponding to the commodity detail page; when the push object is an application detail page, the target operation may be an operation of activating an application corresponding to the application detail page, or the target operation is an operation of downloading the application corresponding to the application detail page, or the target operation is an operation of transferring electronic resources in order to obtain a usage right of the application corresponding to the application detail page, or the like; the electronic resources include funds such as french currency, electronic money, and the like for purchasing usage rights of goods or applications; the electronic money refers to money stored in an electronic form in an electronic wallet (e.g., QQ wallet, wechat wallet, etc.) held by an account, and may include, but is not limited to, electronic bills, digital money (an unregulated, digitized money), and the like.
A terminal: may be a mobile terminal, a fixed terminal, or a portable terminal such as a mobile handset, station, unit, device, multimedia computer, multimedia tablet, internet node, communicator, desktop computer, laptop computer, notebook computer, netbook computer, tablet computer, Personal Communication System (PCS) device, personal navigation device, Personal Digital Assistant (PDA), audio/video player, digital camera/camcorder, positioning device, television receiver, radio broadcast receiver, electronic book device, game device, or any combination thereof, including accessories and peripherals of these devices, or any combination thereof.
The following explains the design concept of the present disclosure.
In the related art, in the process of predicting whether a user performs a specified operation on a push object, generally, historical data including user operation characteristic data and an operation result of whether the user performs the specified operation is obtained, a data association relation between the user operation characteristic data and the operation result in the obtained historical data is analyzed, and whether the user to be predicted performs the specified operation on the push object is estimated based on the analysis result and the user operation characteristic data of the user to be predicted.
For example, in internet e-commerce, e-commerce type advertisements are an important category, and the purpose of the type of advertisements is to guide users to enter a commodity detail page linked with commodities of an advertiser, complete a specified operation of purchasing corresponding commodities and promote the sale of the corresponding commodities of the advertiser; the E-commerce advertisement generally comprises a traffic party, an advertisement platform and a plurality of participants of an advertiser, wherein the traffic party provides traffic service required in the whole process, the advertisement platform provides the platform of the E-commerce advertisement, the advertiser can push own commodities to users in a commodity link form through the advertisement platform, and the users can click the commodity link to enter a commodity detail page to purchase corresponding commodities; at present, e-commerce type advertisement is generally realized through an information flow bidding advertisement cpa (Cost Per action) or ocpc (optimized Cost Per click) mechanism, in this way, an advertiser sets up an optimization target (i.e. the above-mentioned specified operation) and a bid (selling price) for the e-commerce type advertisement, and the optimization target is used as a basis for bidding for an advertisement platform, i.e. the core logic of the method is to calculate expected revenue ecpm of the advertiser based on the following formula 1, and the advertisement platform is more likely to win from the bid the higher the ecpm is, so as to bid for the traffic of the advertisement; in this scenario, the advertisement platform usually estimates the click rate ctr and the conversion rate cvr of a user clicking a certain advertisement (i.e. the above commodity link) through machine learning and deep learning techniques, and the more accurate estimation, the greater the revenue brought to the advertiser and the advertisement platform.
Equation 1: ecpm ═ cpa bid × cvr × ctr;
ecpm in the formula 1 is expected income, and represents income brought to an advertiser by pushing commodity links for a set number of times on an advertisement platform; ctr is click rate and represents the probability of clicking the pushed commodity link by the user; cvr is conversion rate, which represents the ratio of the number of times that the user purchases the corresponding commodity of the commodity link to the number of times that the commodity link is pushed; cpa _ bid is the bid (selling price) of the optimization objective, cpa _ bid may be 100 yuan when the optimization objective is selling goods, that is, the advertiser bids 100 yuan for one of the goods, cpa _ bid may be 40 yuan when the optimization objective is activating the application, that is, the advertiser bids 40 yuan for one of the user's operations of activating the application, and so on.
When the click rate ctr and the conversion rate cvr are estimated by using a machine learning technology, historical click data and conversion data related to a pushed object are needed to finish training and online service of a machine learning model in machine learning; at present, some flow parties and advertisement platforms are integrated, namely some flow parties build advertisement platforms by themselves, if an advertiser carries out E-commerce advertisements on the advertisement platform, a data acquisition link of corresponding click data (such as data of clicking commodity links to enter commodity detail pages) and conversion data (such as data of adding commodities corresponding to the commodity links to shopping lists and data of purchasing commodities corresponding to the commodity links) is smooth, the advertisement platform can directly acquire historical click data and conversion data, a machine learning model is trained on the basis of the historical click data and the conversion data to complete estimation of commodity conversion rate cvr (namely probability of purchasing commodities corresponding to the commodity links pushed by a user), and then the advertisement platform can provide the estimated conversion rate as an optimization target for the advertiser to select.
If the traffic party and the advertisement platform are not the same platform, the data acquisition link of the conversion data of the e-commerce advertisement is limited, the advertisement platform cannot acquire the historical conversion data, and further cannot train the machine learning model to complete the estimation of the commodity conversion rate cvr based on the historical conversion data; similarly, if the optimization target of the advertiser is to activate the application, the advertisement platform cannot acquire the conversion rate of the application activated by the user (i.e., the ratio of the number of times that the user activates the application to the number of times that the advertisement platform pushes the application), and further cannot estimate the conversion rate of the application activation based on machine learning, so that how to predict the operation of the account is a problem to be considered when historical data including a real operation result of the operation of the account is not obtained.
In view of this, the present disclosure designs a target operation prediction method, apparatus, electronic device and storage medium, for optimizing a prediction process of a target operation; considering whether the account performs the target operation on the push object (i.e. the above-mentioned designated operation on the push object) is related to the interest level of the account in the push object, generally, the higher the interest level of the account in the push object is, the more likely the account performs the target operation on the push object, and the information of the account browsing the push object can reflect the interest level of the account in the push object to a great extent. Therefore, under the condition that historical browsing object data containing the operation result of the historical account for performing the target operation on the push object is not obtained, the embodiment of the application considers that the interest degree of the historical account for the push object is determined according to the information of the historical account browsing push object, and further determines whether the historical account performs the approximate target operation result of the target operation or not according to the interest degree of the historical account for the push object; and then, estimating a second degree of association between the operation intention characteristic data of the account to be predicted and the target operation result by utilizing the first degree of association between the operation intention characteristic data of the historical account (namely the user operation characteristic data) and the approximate target operation result of the historical account, and converting the estimated second degree of association into an operation intention probability value for the account to be predicted to perform the target operation.
In the information of the account browsing push object, the browsing duration can reflect the preference degree of the account to the push object to a greater extent, so that when the approximate target operation result of the historical account is determined according to the information of the historical account browsing push object, the approximate target operation result of the historical account can be determined according to the duration information of the historical account browsing push object.
The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings:
as shown in fig. 1, an embodiment of the present disclosure provides an application scenario, where the application scenario includes at least one terminal 101 and a server 102, where:
the server 102 is used for sending a push object to an account logged by a terminal, acquiring operation intention characteristic data of an account to be predicted, and estimating a second association degree between the operation intention characteristic data of the account to be predicted and a target operation result according to the operation intention characteristic data of each historical account and a first association degree of the target operation result determined by time length data of the push object browsed by each historical account; and converting the estimated second relevance into an operation intention probability value of the account to be predicted for carrying out the target operation.
The terminal 101 is configured to log in an account according to an instruction of the user, browse a push object sent by the server 102 according to the instruction of the user, or perform a target operation after browsing the push object according to the instruction of the user.
As shown in fig. 2, an embodiment of the present disclosure provides a target operation prediction method, which specifically includes the following steps:
step S201, obtaining operation intention characteristic data of the account to be predicted, wherein the operation intention characteristic data comprises characteristic data representing the preference degree of the account to target operation, and the target operation comprises operation performed after the account browses a push object.
As an embodiment, the operation intention may include at least one feature data of an account feature of the account to be predicted, a feature of the push object, and a scene feature of the push object viewed by the account to be predicted.
The account characteristics may include, but are not limited to, one or more information of user profile characteristics corresponding to the account, time of account creation, history of target operation performed by the account on the push object, and the like, where the user profile characteristics may include age, gender, height, weight, native place, nationality, current geographic area, favorite object type, occupation, hobby, and the like of the user.
When the push object is an advertisement object of the internet e-commerce platform, the characteristics of the push object may include, but are not limited to, one or more of the advertisement category of the push object in the internet advertisement platform, the time period of the push object being pushed in the internet advertisement platform, and the advertiser characteristics of the advertiser to which the push object belongs, wherein the advertiser characteristics may be identification information of the advertiser, historical sales information of the advertiser, and the like.
The scene characteristics may include, but are not limited to, one or more of a network connection mode of a terminal that logs in an account to be predicted, a type of the terminal, a current geographic area of the terminal, and a current time when the account to be predicted browses the push object, where the network connection mode may include, but is not limited to, a wired network connection or a wireless network connection such as a wifi connection, and the like, and the type of the terminal may be a mobile terminal such as a smart phone, a laptop, a multimedia tablet, and the like; but may also be a fixed terminal such as a website, desktop computer, internet node, etc.
Step S202, estimating a second degree of association between the operation intention characteristic data of the account to be predicted and a target operation result according to the operation intention characteristic data of each historical account and the first degree of association of the target operation result determined by the duration data of the pushed object browsed by each historical account; and converting the estimated second relevance into an operation intention probability value of the account to be predicted for carrying out the target operation, wherein the target operation result comprises the result of whether the target operation is carried out or not.
As an embodiment, in the embodiment of the present disclosure, a first degree of association between the operation intention characteristic data of each historical account and a target operation result determined by the duration data of browsing the push object by each historical account may be learned through machine learning, a second degree of association between the operation intention characteristic data of the account to be predicted and the target operation result is calculated based on the learned first degree of association, and the estimated second degree of association is converted into an operation intention probability value for performing the target operation; the first degree of association may be learned by, but is not limited to, machine learning methods such as supervised classification learning, unsupervised classification learning, reinforcement learning, and deep reinforcement learning.
The operation intention probability value in the embodiment of the disclosure may be a value between 0 and 1, and the value of the operation intention probability value represents a probability value of a target operation performed after the account to be tested browses the push object.
The following disclosure provides a process of estimating a probability value of an operation intention of the account to be predicted to perform the target operation in step S202 by using a machine learning method, specifically as follows:
before the probability value of the operation intention of the account to be predicted for carrying out the target operation is estimated, the operation intention characteristic data of each historical account and the time length data of the objects pushed by the historical accounts in a browsing mode can be used as training samples on the basis of a machine learning method, and an operation intention estimation model is obtained through training.
And then when the operation intention probability value of the account to be predicted for carrying out the target operation is estimated, inputting the extracted operation intention characteristic data of the account to be predicted by adopting the trained operation intention estimation model, and obtaining the operation intention probability value of the account to be predicted for carrying out the target operation, which is output by the operation intention estimation model.
The embodiment of the disclosure provides a training process of an operation intention estimation model, which specifically includes the following steps:
the method comprises the steps of firstly, acquiring historical browsing object data as a training sample, wherein the historical browsing object data comprises duration data of a historical account browsing recommended object and operation intention characteristic data of the historical account for performing target operation.
Further, historical browsing object data within a set time length before the current time may be determined as training samples, or historical browsing object data of a second set number before the current time may be determined as training samples.
As an example, the acquired historical browsing object data may have a problem of data missing, such as missing of a certain item or items of data in the operation intention characteristic data, in which case the missing data in the historical browsing object data may be processed by setting a character string.
Training an operation intention estimation model by using the obtained training sample; specifically, an initially created operational intention prediction model may be trained based on initially obtained training samples; or after the trained operation intention estimation model is obtained in the history training, the operation intention model is updated according to the newly generated browsing data of the history object, for example, the training sample is updated according to the newly generated browsing data of the history object, and the operation intention estimation model obtained by the history training is adjusted by using the operation intention characteristic data of the history account in the updated training sample and the target operation result determined by the duration data of the history account browsing the push object, so as to obtain the new trained operation estimation model.
The process of obtaining the trained predictive model of operational intent is described in detail below.
The first model acquisition mode: and adjusting the operation intention estimation model initially created through machine learning.
As shown in fig. 3, the training process of the operation intention estimation model in this manner mainly includes the following steps:
step S301, a training sample and an initially created operation intention estimation model are obtained.
The initially created operation intention prediction model may be a supervised machine learning model, such as but not limited to an xgboost model, a deep learning model, a logistic regression model, etc.
In this step, each training sample includes operation intention characteristic data of the historical account and duration data of browsing the push object by the historical account, and the training sample may be historical browsing object data acquired from a designated platform, such as an internet e-commerce platform, or may be acquired from the historical browsing object data according to other manners, and a person skilled in the art may acquire the training sample in this step according to actual needs.
The training sample in this step may be historical browsing data within a set time period before the current time, such as historical browsing data within one month before the current time; the training sample may also be a set amount of historical browsing data before the current time, such as 100 ten thousand historical browsing data before the current time, where the current time may be understood as a time at which an initial operation intention estimation model is created or a time at which the training of the operation intention estimation model is triggered.
Step S302, extracting the operation intention characteristic data of the historical account and the duration data of the historical account browsing the push object in each training sample.
Since the operation intention characteristic data in the embodiment of the present disclosure may include a plurality of characteristic data, in this step, the training sample may be processed by setting a processing mode, and the operation intention characteristic data of the historical account and the duration data of the historical account browsing and pushing the object are extracted.
Step S303, determining a target operation result of the historical account in each training sample according to the duration data of the pushing object browsed by the historical account in each training sample.
In general, the longer the time for browsing the push object by the account is, the higher the possibility that the account performs the target operation after browsing the push object is, and if the time for browsing the commodity detail page corresponding to the commodity link by the account is longer, the higher the possibility that the account purchases the commodity corresponding to the commodity detail page is; and the shorter the time for browsing the push object by the account is, the lower the possibility that the account performs the target operation after browsing the push object is, so that the target operation result of performing the target operation on the historical account in the training sample can be determined based on the duration data for browsing the push object in the training sample, where the target operation result includes performing the target operation or not performing the target operation.
Further, in order to facilitate learning of the first association degree between the operation intention characteristic data of the historical account in the training samples and the target operation result of the historical account, the target operation result of the historical account in each training sample may be labeled by a label, for example, the label of the training sample with the target operation result being performed is labeled as 1, the label of the training sample with the target operation result being not performed is labeled as 0, and the like, that is, the label of each training sample may be determined by browsing the duration data of the pushed object through the historical account in each training sample.
Step S304, the initially created operation intention estimation model is adjusted through machine learning by utilizing the mapping relation between the operation intention characteristic data of the historical account in each training sample and the corresponding target operation result, and the trained operation intention estimation model is obtained.
Inputting the operation intention characteristic data of the historical account and the target operation result in the training sample into an initially created operation intention estimation model, learning the first association degree of the operation intention characteristic data of the historical account and the target operation result in the training sample by the initially created operation intention estimation model, and adjusting parameters in the initially created operation intention estimation model to obtain the trained operation intention estimation model.
When the operation intention characteristic data and the target operation result of the historical account in the training sample are input into the initially created intention operation estimation model, the operation intention characteristic data and the target operation result of the historical account in the training sample can be input into the initially created operation intention estimation model in batches, for example, the operation intention characteristic data and the target operation result of the historical account in a set number of training samples are input at one time, or the operation intention characteristic data and the target operation result of the historical account in the training sample in one time period are input at one time, and the like.
If the target operation result corresponding to the training sample is embodied in the form of the label, the operation intention characteristic data of the historical account in the training sample and the label can be input into the initially created operation intention estimation model for training, and the trained operation intention estimation model is obtained.
After the initially created operation intention estimation model is adjusted to the trained operation intention estimation model, the trained operation intention estimation model can be used for predicting the target operation of the account to be predicted, and in the process of predicting the target operation of the account to be predicted by using the trained machine learning model, the trained operation intention estimation model can be used as an operation intention estimation model obtained through historical training and retrained to obtain a new trained operation intention estimation model, for example, the operation intention estimation model obtained through historical training is periodically adjusted in a set time period.
The second model acquisition mode: and training the operation intention estimation model obtained by historical training to obtain a new trained operation intention estimation model.
As shown in fig. 4, the training process of the operation intention estimation model in this manner mainly includes the following steps:
step S401, updating a training sample based on historical browsing object data before the current time;
it should be noted that, the current time here may be understood as a time for triggering the update of the operation intention estimation model, for example, a time interval between the current time and a time at which the last trained machine learning model is obtained reaches the set time period, or a time interval between the current time and a time at which the last trained operation intention estimation model is triggered reaches the set time period, or a time at which the update of the operation intention estimation model is triggered due to some reason, such as service upgrade, etc.
Specifically, the historical browsing object data within a set time period before the current time may be determined as the training sample, and for example, the historical browsing object data generated between the current time and the time when the training operation intention estimation model is triggered last time is determined as the training sample; the historical browsing object data of a second set number before the current time can be determined as training samples; the historical browsing object data in a set time length before the current time and the training samples used for obtaining the estimation model of the operation intention of the last training can be determined as the training samples of the step, for example, the training samples used for obtaining the estimation model of the operation intention of the last training and the historical browsing object data generated between the current time and the time of obtaining the estimation model of the operation intention of the last triggering training are determined as the training samples.
Step S402, determining the target operation result of the historical account in each training sample according to the updated duration data of the historical account browsing object in each training sample.
This step can be referred to the description of step S303, and will not be repeated here.
And step S403, adjusting an operation intention estimation model obtained by historical training through machine learning by utilizing the mapping relation between the operation intention characteristic data of the historical account in each training sample and the corresponding target operation result, and obtaining the operation intention estimation model after the training.
The operation intention characteristic data of the historical account in the training sample and the corresponding target operation result can be input into the machine learning model of the historical training, the machine learning model of the historical training learns the mapping relation between the operation intention characteristic data of the historical account in the training sample and the target operation result, parameters in the operation intention estimation model obtained through the historical training are adjusted, and the operation intention estimation model after the training is obtained.
The process of inputting the operation intention characteristic data and the target operation result in the training sample into the operation intention estimation model obtained by historical training may be referred to the description in step S304, and will not be repeated here.
As an embodiment, in steps S303 and S402, the target operation result of the historical object in each training sample may be determined according to the duration data of browsing the pushed object by the historical account in each training sample in the following manners.
The first operation result determination mode:
the method comprises the steps that a target operation result corresponding to a training sample with the duration of a historical account browsing a pushed object not less than a set duration threshold is determined as a first result of the target operation; and
and determining a target operation result corresponding to the training sample with the duration of the historical account browsing the pushed object being less than the set duration threshold as a second result of not performing the target operation.
If the target operation result is represented in the form of a label, the first result may be set as a positive sample label and the second result may be set as a negative sample label.
In this way, for a training sample without duration data of a historical account browsing a pushed object, the duration of the browsing object of the historical account can be regarded as zero, and further, when the target operation result of such a training sample is the second result, that is, when the pushed object is a commodity detail page corresponding to a commodity link, the target operation result of the training sample without the historical account browsing the commodity detail page is regarded as the second result.
The second operation result determination mode:
the method comprises the steps that a target operation result corresponding to a training sample with the duration of a historical account browsing a pushed object not less than a set duration threshold is determined as a first result of the target operation; and
determining a target operation result corresponding to the training sample with the duration of the historical account browsing the pushed object being less than the set duration threshold as a second result of not performing the target operation; and
and determining a target operation result corresponding to the training sample without the duration data of the historical account browsing push object as a second result without the target operation.
If the target operation result is represented in the form of a label, the first result may be set as a positive sample label and the second result may be set as a negative sample label.
The positive sample label in the first operation result determining manner and the negative sample label in the second operation result determining manner may be set to 1, and the negative sample label may be set to 0.
As an embodiment, in the first operation result determining manner and the second operation result determining manner, the set duration threshold is determined according to duration data of a first set number of history objects browsing the pushed objects and actual target operation results of the history objects, or the set duration threshold is obtained through an AB test.
Specifically, a first set amount of historical browsing object data may be obtained in advance, where the first set amount of historical browsing object data includes duration data of a pushing object browsed by a historical account and a real target operation result of the historical account.
Referring to fig. 5, a specific example of obtaining the set duration threshold is provided as follows, which specifically includes the following steps:
it should be noted that each of the historical browsing object data in this example is among the above-described first set number of historical browsing data.
In step S501, an initial time threshold T0 is set.
Specifically, the time period threshold T0 may be set by a skilled person according to experimental tests or empirical estimation.
Step S502, according to the time length threshold T0 and the time length data of the pushing object browsed by the historical account in the historical browsing object data, determining the approximate target operation result of the historical account in the historical browsing object data.
Specifically, the approximate target operation result includes that a target operation is performed or the target operation is not performed, and if the duration of browsing the push object by the historical account in the historical browsing object data is not less than the duration threshold T0, the approximate target operation result in the historical browsing object data is determined to be that the target operation is performed; and if the time length for browsing the push object by the historical account in the historical browsing object data is less than a time length threshold T0, determining that the approximate target operation result of the historical object in the historical browsing object data is that no target operation is performed.
In step S503, the prediction error rate of the target operation result is determined based on the actual target operation result and the approximate target operation result of each history browsing object data.
Specifically, the number of pieces of history browsing object data in which the approximated target operation result and the true target operation result do not coincide is determined, and the ratio of the determined number to the first set number is used as the prediction error rate of the target operation result.
Step S504, determining whether the predicted error rate is greater than the error rate threshold, if so, going to step S505, and if not, going to step S506.
In step S505, the duration threshold T0 is adjusted to obtain a new duration threshold T0, and the process goes to step S502.
In this step, the product of the current time length threshold T0 and the set weight may be determined as the new time length threshold T0; the sum of the current duration threshold T0 and the set duration may also be determined as the new duration threshold T0, or the current duration threshold T0 may be transformed in other ways to obtain a new duration threshold T0.
In step S506, the duration threshold T0 is determined as the set duration threshold.
One specific example of target operation prediction is provided below.
In the example, the pushed object is a commodity detail page linked with a commodity, the target operation is an operation of transferring electronic money for the commodity corresponding to the commodity detail page (namely an operation of purchasing the commodity corresponding to the commodity detail page), the time length information of the account browsing the pushed object is the time length of the account browsing the commodity detail page, and the probability value of the operation intention of the account browsing the pushed object to be predicted is estimated by adopting a trained operation intention estimation model.
The following two processes are mainly included in this example:
the first process is as follows: and acquiring a trained operation intention estimation model.
Firstly, acquiring a training sample containing operation intention characteristic data of a historical account and duration information of a pushing object browsed by the historical account; setting the label of the training sample with the duration of the historical account for browsing the pushed object not less than the set duration threshold as a positive sample label, setting the label of the training sample with the duration of the historical account for browsing the pushed object less than the set duration threshold as a negative sample label, the positive sample tag is used for representing that the historical account transfers electronic money for the commodities corresponding to the browsed commodity detail page (namely the commodities corresponding to the commodity detail page are purchased after the commodity detail page is browsed by the historical account), and the negative sample tag is used for representing that the electronic money is not transferred for the commodities corresponding to the commodity detail page by the historical account (namely the commodities corresponding to the commodity detail page are not purchased after the commodity detail page is browsed by the historical account, or the commodities corresponding to the commodity detail page are not purchased after the commodities are browsed by the historical account).
Based on machine learning, the initially created operation intention estimation model or the operation intention estimation model obtained by historical training is adjusted by learning the mapping relation between the operation intention characteristic data of the historical account in the training sample and the label, and the trained operation intention estimation model is obtained.
The second process: and estimating the operation intention probability value of the account to be predicted for carrying out target operation.
And acquiring the operation intention characteristic data of the account to be tested, inputting the acquired operation intention characteristic data into the trained operation intention estimation model, and acquiring the operation intention probability value of target operation after the account to be tested browses the push object.
In the embodiment of the disclosure, an operation intention pre-evaluation value of the target operation of the account to be predicted after the push object is browsed is estimated by using the operation intention characteristic data of the historical account and the first relevance of the target operation result determined by the time length data of the push object browsed by each historical account; in the process, the real target operation result of the target operation performed by the historical account does not need to be acquired, and a new target operation prediction mode is provided.
As shown in fig. 6, based on the same inventive concept, the disclosed embodiment further provides a target operation prediction apparatus 600, which includes:
an intention characteristic obtaining unit 601 configured to perform obtaining of operation intention characteristic data of an account to be predicted, where the operation intention characteristic data includes characteristic data representing a degree of preference of the account for a target operation, and the target operation includes an operation performed after the account browses a push object;
a target operation prediction unit 602 configured to perform a first degree of association of a target operation result determined by operation intention characteristic data of each historical account and duration data of browsing a push object by each historical account, and estimate a second degree of association of the operation intention characteristic data of the account to be predicted and the target operation result; and converting the estimated second degree of association into an operation intention probability value of the account to be predicted for performing the target operation, wherein the target operation result comprises a result of whether the target operation is performed or not.
As an embodiment, the target operation prediction unit 602 is specifically configured to perform:
inputting the operation intention characteristic data of the account to be predicted by adopting a trained operation intention prediction model, and obtaining the operation intention probability value of the account to be predicted for carrying out the target operation, which is output by the operation intention prediction model, wherein the operation intention prediction model is obtained by training by adopting the operation intention characteristic data of each historical account and the time length data of the object to be pushed and browsed by each historical account as training samples based on a machine learning method.
As an embodiment, the target operation prediction unit 602 is further configured to perform:
the trained operation intention estimation model is obtained by the following method:
extracting operation intention characteristic data of the historical account and duration data of the historical account browsing and pushing objects in each training sample;
according to the duration data of the pushed object browsed by the historical account in each training sample, determining a target operation result corresponding to each training sample;
and obtaining the trained operation intention estimation model by utilizing the mapping relation between the operation intention characteristic data of the historical account in each training sample and the corresponding target operation result and adjusting the operation intention estimation model obtained by historical training or the initially created operation intention estimation model through machine learning.
As an embodiment, the target operation prediction unit 602 is specifically configured to perform:
according to the time length data of the historical account browsing the pushed object in each training sample, determining the target operation result corresponding to the training sample with the time length of the historical account browsing the pushed object not less than the set time length threshold value as the first result of the target operation, and
and determining a target operation result corresponding to the training sample with the duration of the historical account browsing the pushed object being less than the set duration threshold as a second result of not performing the target operation.
As an embodiment, the target operation prediction unit 602 is further configured to perform:
and determining a target operation result corresponding to the training sample without the duration data of the historical account browsing push object as a second result without the target operation.
As an embodiment, the set duration threshold is determined according to duration data of a first set number of history objects browsing the push object and corresponding real target operation results.
As an embodiment, the target operation prediction unit 602 is further configured to perform:
determining historical browsing object data within a set time length before the current time as the training sample; or
And determining the historical browsing object data of a second set number before the current time as the training sample.
As shown in fig. 7, the present disclosure provides an electronic device 700 comprising a processor 701, a memory 702 for storing the processor-executable instructions described above;
wherein the processor is configured to execute executable instructions to implement any of the target operation prediction methods described above.
In an exemplary embodiment, a storage medium comprising instructions, such as a memory comprising instructions, executable by a processor of the electronic device to perform the method is also provided. Alternatively, the storage medium may be a non-transitory computer readable storage medium, for example, which may be a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims (10)
1. A method of target operation prediction, comprising:
acquiring operation intention characteristic data of an account to be predicted, wherein the operation intention characteristic data comprises characteristic data representing the preference degree of the account on target operation, and the target operation comprises operation performed after the account browses a push object;
estimating a second degree of association between the operation intention characteristic data of the account to be predicted and a target operation result according to the first degree of association of the operation intention characteristic data of each historical account and the target operation result determined by the duration data of the pushing object browsed by each historical account; and converting the estimated second relevance into an operation intention probability value of the account to be predicted for performing the target operation, wherein the target operation result comprises a result of whether the target operation is performed or not.
2. The method according to claim 1, wherein the second association degree of the operation intention characteristic data of the account to be predicted with the target operation result is estimated through the first association degree of the operation intention characteristic data of each historical account and the target operation result determined by the duration data of the pushing object browsed by each historical account; and converting the estimated second relevance into an operation intention probability value of the account to be predicted for performing the target operation, wherein the step comprises the following steps of:
inputting the operation intention characteristic data of the account to be predicted by adopting a trained operation intention prediction model, and obtaining the operation intention probability value of the account to be predicted for carrying out the target operation, which is output by the operation intention prediction model, wherein the operation intention prediction model is obtained by training by adopting the operation intention characteristic data of each historical account and the time length data of the object to be pushed browsed and pushed by each historical account as training samples based on a machine learning method.
3. The method of claim 2, wherein the trained predictive model of operational intent is obtained by:
extracting operation intention characteristic data of the historical account and duration data of the historical account browsing and pushing objects in each training sample;
according to the duration data of the pushed object browsed by the historical account in each training sample, determining a target operation result corresponding to each training sample;
and obtaining the trained operation intention estimation model by utilizing the mapping relation between the operation intention characteristic data of the historical account in each training sample and the corresponding target operation result and adjusting the operation intention estimation model obtained by historical training or the initially created operation intention estimation model through machine learning.
4. The method of claim 3, wherein the step of determining the target operation result of the historical object in each training sample according to the duration data of the historical account browsing the pushed object in each training sample comprises:
according to the time length data of the historical account browsing the pushed object in each training sample, determining the target operation result corresponding to the training sample with the time length of the historical account browsing the pushed object not less than the set time length threshold value as the first result of the target operation, and
and determining a target operation result corresponding to the training sample with the duration of the historical account browsing the pushed object being less than the set duration threshold as a second result of not performing the target operation.
5. The method of claim 4, further comprising:
and determining a target operation result corresponding to a training sample without the duration data of the historical account browsing push object as a second result without the target operation.
6. The method of claim 4, wherein the set duration threshold is determined according to duration data of a first set number of historical objects browsing the pushed object and corresponding real target operation results.
7. The method of claim 3, further comprising:
determining historical browsing object data within a set time length before the current time as the training sample; or
And determining a second set amount of historical browsing object data before the current time as the training sample.
8. A target operation prediction apparatus, comprising:
the intention characteristic obtaining unit is configured to execute obtaining operation intention characteristic data of the account to be predicted, the operation intention characteristic data comprise characteristic data representing the preference degree of the account on target operation, and the target operation comprises operation performed after the account browses a push object;
the target operation prediction unit is configured to execute a first association degree of a target operation result determined by operation intention characteristic data of each historical account and duration data of browsing a push object by each historical account, and estimate a second association degree of the operation intention characteristic data of the account to be predicted and the target operation result; and converting the estimated second relevance into an operation intention probability value of the account to be predicted for performing the target operation, wherein the target operation result comprises a result of whether the target operation is performed or not.
9. An electronic device comprising a processor, a memory for storing instructions executable by the processor;
wherein the processor is configured to perform the method of any one of claims 1 to 7.
10. A computer-readable storage medium having stored thereon computer instructions which, when executed on a computer, cause the computer to perform the method of any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010469618.8A CN113743636A (en) | 2020-05-28 | 2020-05-28 | Target operation prediction method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010469618.8A CN113743636A (en) | 2020-05-28 | 2020-05-28 | Target operation prediction method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113743636A true CN113743636A (en) | 2021-12-03 |
Family
ID=78724227
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010469618.8A Pending CN113743636A (en) | 2020-05-28 | 2020-05-28 | Target operation prediction method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113743636A (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107360222A (en) * | 2017-06-30 | 2017-11-17 | 广东欧珀移动通信有限公司 | Merchandise news method for pushing, device, storage medium and server |
| CN107742250A (en) * | 2017-12-04 | 2018-02-27 | 深圳春沐源控股有限公司 | The display methods and display system of goods browse record |
| CN108665329A (en) * | 2017-03-29 | 2018-10-16 | 北京京东尚科信息技术有限公司 | A kind of Method of Commodity Recommendation based on user browsing behavior |
| CN109165983A (en) * | 2018-09-04 | 2019-01-08 | 中国平安人寿保险股份有限公司 | Insurance products recommended method, device, computer equipment and storage medium |
| CN110020176A (en) * | 2017-12-29 | 2019-07-16 | 中国移动通信集团公司 | A kind of resource recommendation method, electronic equipment and computer readable storage medium |
| CN110716979A (en) * | 2019-10-18 | 2020-01-21 | 重庆锐云科技有限公司 | House buying intention client mining method, device and server |
| CN110929206A (en) * | 2019-11-20 | 2020-03-27 | 腾讯科技（深圳）有限公司 | Click rate estimation method and device, computer readable storage medium and equipment |
| CN111177575A (en) * | 2020-04-07 | 2020-05-19 | 腾讯科技（深圳）有限公司 | Content recommendation method and device, electronic equipment and storage medium |
-
2020
- 2020-05-28 CN CN202010469618.8A patent/CN113743636A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108665329A (en) * | 2017-03-29 | 2018-10-16 | 北京京东尚科信息技术有限公司 | A kind of Method of Commodity Recommendation based on user browsing behavior |
| CN107360222A (en) * | 2017-06-30 | 2017-11-17 | 广东欧珀移动通信有限公司 | Merchandise news method for pushing, device, storage medium and server |
| CN107742250A (en) * | 2017-12-04 | 2018-02-27 | 深圳春沐源控股有限公司 | The display methods and display system of goods browse record |
| CN110020176A (en) * | 2017-12-29 | 2019-07-16 | 中国移动通信集团公司 | A kind of resource recommendation method, electronic equipment and computer readable storage medium |
| CN109165983A (en) * | 2018-09-04 | 2019-01-08 | 中国平安人寿保险股份有限公司 | Insurance products recommended method, device, computer equipment and storage medium |
| CN110716979A (en) * | 2019-10-18 | 2020-01-21 | 重庆锐云科技有限公司 | House buying intention client mining method, device and server |
| CN110929206A (en) * | 2019-11-20 | 2020-03-27 | 腾讯科技（深圳）有限公司 | Click rate estimation method and device, computer readable storage medium and equipment |
| CN111177575A (en) * | 2020-04-07 | 2020-05-19 | 腾讯科技（深圳）有限公司 | Content recommendation method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20180165745A1 (en) | Intelligent Recommendation Method and System | |
| CN109783730A (en) | Products Show method, apparatus, computer equipment and storage medium | |
| US20220012768A1 (en) | Iteratively improving an advertisement response model | |
| CN105160545B (en) | Method and device for determining release information style | |
| CN110008973B (en) | Model training method, method and device for determining target user based on model | |
| CN108074003B (en) | Prediction information pushing method and device | |
| CN110598120A (en) | Behavior data based financing recommendation method, device and equipment | |
| CN111429214B (en) | Transaction data-based buyer and seller matching method and device | |
| CN111667024B (en) | Content pushing method, device, computer equipment and storage medium | |
| CN113761348A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN111651679A (en) | Recommendation method and device based on reinforcement learning | |
| CN113781149A (en) | Information recommendation method and device, computer-readable storage medium and electronic equipment | |
| CN114862480A (en) | Advertisement putting orientation method and its device, equipment, medium and product | |
| CN111782937A (en) | Information sorting method and device, electronic equipment and computer readable medium | |
| US20230316106A1 (en) | Method and apparatus for training content recommendation model, device, and storage medium | |
| US20220198487A1 (en) | Method and device for processing user interaction information | |
| CN111787042B (en) | Method and device for pushing information | |
| CN116911953A (en) | Article recommendation method, apparatus, electronic device and computer readable storage medium | |
| CN113034168A (en) | Content item delivery method and device, computer equipment and storage medium | |
| CN113360816A (en) | Click rate prediction method and device | |
| US11776011B2 (en) | Methods and apparatus for improving the selection of advertising | |
| CN113743636A (en) | Target operation prediction method and device, electronic equipment and storage medium | |
| CN114971716A (en) | Service interface quality evaluation method and device, equipment, medium and product thereof | |
| CN113822734A (en) | Method and apparatus for generating information | |
| CN110348947B (en) | Object recommendation method and device |
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 |