CN111461778B - Advertisement pushing method and device - Google Patents
Advertisement pushing method and device Download PDFInfo
- Publication number
- CN111461778B CN111461778B CN202010242625.4A CN202010242625A CN111461778B CN 111461778 B CN111461778 B CN 111461778B CN 202010242625 A CN202010242625 A CN 202010242625A CN 111461778 B CN111461778 B CN 111461778B
- Authority
- CN
- China
- Prior art keywords
- advertisement
- model
- time
- user
- reading
- 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.)
- Active
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
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application provides a method and a device for pushing advertisements, which acquire advertisement information of advertisements to be published and user information of target users; processing advertisement information of advertisements to be released and user information of target users by utilizing a pre-trained push time model to obtain push time of the advertisements to be released, which corresponds to the target users; the push time model is a neural network model obtained by training a plurality of training samples; each training sample comprises user information of a historical user, advertisement information of a published advertisement which is effectively read by the historical user and actual reading time; and pushing the advertisement to be released to the target user at the advertisement pushing time to be released corresponding to the target user. The pushing time model provided by the scheme can be used for predicting the time when the user is most likely to effectively read the advertisement to be published according to the preferences of the user for different advertisements in different time periods in the past, and pushing the advertisement in the time, so that the advertisement pushing efficiency is further improved.
Description
Technical Field
The invention relates to the technical field of advertisements, in particular to a method and a device for pushing advertisements.
Background
With the development of computer technology and internet technology, the use range of various forms of computers and smartphones is becoming wider and wider. Accordingly, pushing advertisements to users while they are using these electronic devices is also a common means of information promotion.
At present, when a user uses a device to access a specific network platform (such as a webpage, client software and various application programs), the advertisement pushing strategy for the electronic device generally determines the preference of the user to information according to the type of information browsed by the user in the past and part of personal information of the user, and then pushes advertisement content related to the preferred information to the user, so that the user is ensured to have higher probability of reading the pushed advertisement, and the effect of improving the advertisement pushing efficiency is achieved.
However, existing methods push the same type of advertisement to the user at any time, without considering that the preference of the user may change in different time periods of the day, and thus the push efficiency of the method is still low.
Disclosure of Invention
Aiming at the problems in the prior art, the application provides a method and a device for pushing advertisements, and particularly provides a scheme for determining the pushing time of the advertisements by analyzing the preference of users in different time periods, so as to further improve the pushing efficiency of the advertisements.
The first aspect of the application discloses an advertisement pushing method, which comprises the following steps:
acquiring advertisement information of an advertisement to be published and user information of a target user;
processing the advertisement information of the advertisement to be distributed and the user information of the target user by utilizing a pre-trained push time model to obtain the push time of the advertisement to be distributed, which corresponds to the target user; the push time model is a neural network model obtained by training a plurality of training samples; each training sample comprises user information of a historical user, and the historical user effectively reads advertisement information of a published advertisement and actual reading time; the actual reading time refers to the time for the historical user to effectively read the published advertisement;
and pushing the advertisement to be distributed to the target user at the advertisement to be distributed pushing time corresponding to the target user.
Optionally, the process of training the push time model includes:
constructing a plurality of training samples by utilizing a plurality of advertisement reading records in an advertisement reading record library;
acquiring a preset neural network model, and determining initial model parameters of the neural network model to obtain an initial push model;
inputting user information and advertisement information of each training sample into the initial pushing model to obtain pushing time of each training sample;
calculating model loss of the initial push model according to the push time of each training sample and the actual reading time of the training sample;
judging whether the model loss meets a preset model convergence condition or not;
if the model loss does not meet the model convergence condition, updating parameters of the initial pushing model according to the model loss, and returning to execute the step of inputting the user information and the advertisement information of each training sample into the initial pushing model to obtain the pushing time of each training sample;
and if the model loss meets the model convergence condition, determining the initial push model as the push time model.
Optionally, each of the advertisement reading records includes: a user identifier of a historical user and an advertisement identifier of a published advertisement, wherein the historical user reads the starting time and the ending time of the published advertisement;
the constructing a plurality of training samples by using a plurality of advertisement reading records in an advertisement reading record library comprises:
calculating a difference value between the starting time of the advertisement reading record and the ending time of the advertisement reading record aiming at each advertisement reading record to obtain the reading time of the advertisement reading record;
judging whether the reading time length of each advertisement reading record is longer than the preset effective reading time length or not according to each advertisement reading record;
for each advertisement reading record, if the reading time of the advertisement reading record is longer than the effective reading time, determining that the advertisement reading record is an effective reading record;
and combining the user information of the historical user of the effective reading records aiming at each effective reading record, the advertisement information of the released advertisement of the effective reading records and the starting time of the effective reading records to obtain a training sample corresponding to the effective reading records.
Optionally, the determining initial model parameters of the neural network model includes:
randomly generating a plurality of initial parameter individuals according to the structure of the neural network model; wherein each of the initial parameter individuals includes a plurality of parameters;
optimizing the plurality of initial parameter individuals by utilizing a genetic algorithm to obtain an optimal parameter individual;
and determining a plurality of parameters in the optimal parameter individuals as initial model parameters of the neural network model.
Optionally, the user information at least includes: user occupation information, user family information and user preference information;
the advertisement information includes at least advertisement types.
A second aspect of the present application provides an advertisement pushing apparatus, including:
the acquisition unit is used for acquiring advertisement information of the advertisement to be distributed and user information of the target user;
the training unit is used for training a neural network model by utilizing a plurality of training samples to obtain a push time model; each training sample comprises user information of a historical user, and the historical user effectively reads advertisement information of a published advertisement and actual reading time; the actual reading time refers to the time for the historical user to effectively read the published advertisement;
the processing unit is used for processing the advertisement information of the advertisement to be distributed and the user information of the target user by utilizing the pushing time model to obtain the pushing time of the advertisement to be distributed, which corresponds to the target user;
and the pushing unit is used for pushing the advertisement to be distributed to the target user at the advertisement to be distributed pushing time corresponding to the target user.
Optionally, the training unit trains a neural network model by using a plurality of training samples, and is specifically configured to:
constructing a plurality of training samples by utilizing a plurality of advertisement reading records in an advertisement reading record library;
acquiring a preset neural network model, and determining initial model parameters of the neural network model to obtain an initial push model;
inputting user information and advertisement information of each training sample into the initial pushing model to obtain pushing time of each training sample;
calculating model loss of the initial push model according to the push time of each training sample and the actual reading time of the training sample;
judging whether the model loss meets a preset model convergence condition or not;
if the model loss does not meet the model convergence condition, updating parameters of the initial pushing model according to the model loss, and returning to execute the step of inputting the user information and the advertisement information of each training sample into the initial pushing model to obtain the pushing time of each training sample;
and if the model loss meets the model convergence condition, determining the initial push model as the push time model.
Optionally, each of the advertisement reading records includes: a user identifier of a historical user and an advertisement identifier of a published advertisement, wherein the historical user reads the starting time and the ending time of the published advertisement;
when the training unit constructs a plurality of training samples by utilizing a plurality of advertisement reading records in the advertisement reading record library, the training unit is specifically used for:
calculating a difference value between the starting time of the advertisement reading record and the ending time of the advertisement reading record aiming at each advertisement reading record to obtain the reading time of the advertisement reading record;
judging whether the reading time length of each advertisement reading record is longer than the preset effective reading time length or not according to each advertisement reading record;
for each advertisement reading record, if the reading time of the advertisement reading record is longer than the effective reading time, determining that the advertisement reading record is an effective reading record;
and combining the user information of the historical user of the effective reading records aiming at each effective reading record, the advertisement information of the released advertisement of the effective reading records and the starting time of the effective reading records to obtain a training sample corresponding to the effective reading records.
Optionally, when the training unit determines initial model parameters of the neural network model, the training unit is specifically configured to:
randomly generating a plurality of initial parameter individuals according to the structure of the neural network model; wherein each of the initial parameter individuals includes a plurality of parameters;
optimizing the plurality of initial parameter individuals by utilizing a genetic algorithm to obtain an optimal parameter individual;
and determining a plurality of parameters in the optimal parameter individuals as initial model parameters of the neural network model.
Optionally, the user information at least includes: user occupation information, user family information and user preference information;
the advertisement information includes at least advertisement types.
The application provides a method and a device for pushing advertisements, which acquire advertisement information of advertisements to be published and user information of target users; processing advertisement information of advertisements to be released and user information of target users by utilizing a pre-trained push time model to obtain push time of the advertisements to be released, which corresponds to the target users; the push time model is a neural network model obtained by training a plurality of training samples; each training sample comprises user information of a historical user, advertisement information of a published advertisement which is effectively read by the historical user and actual reading time; and pushing the advertisement to be released to the target user at the advertisement pushing time to be released corresponding to the target user. The pushing time model provided by the scheme can be used for predicting the time when the user is most likely to effectively read the advertisement to be published according to the preferences of the user for different advertisements in different time periods in the past, and pushing the advertisement in the time, so that the advertisement pushing efficiency is further improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
Fig. 1 is a flowchart of a method for pushing an advertisement according to an embodiment of the present application;
FIG. 2 is a flowchart of a method for training a push time model according to an embodiment of the present application;
fig. 3 is a schematic structural diagram of an advertisement pushing device according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The advertisement pushing efficiency can be understood as the proportion of the number of times that the user effectively reads the pushed advertisement to the total advertisement pushing number of times in a period of time. For example, assuming that a certain advertisement pushing party in the month of january pushes advertisements to 100 users in total, each user pushes 20 times, the total advertisement pushing number in the month of january is 2000, wherein the user effectively reads 1000 times, and the pushing efficiency of the advertisement in the month of january is 50%. After receiving the prompt information of advertisement pushing, the user enters a corresponding advertisement detailed page and stays for a long enough time (specifically, 15 seconds or more), and the user is considered to effectively read the advertisement.
The existing advertisement pushing method only considers what type of users may be interested in what type of advertisements, but does not consider that the preference of the advertisement types of the users may also change with different time periods, so that the advertisement pushing efficiency is not high. Aiming at the problem, the application provides a method and a device for pushing advertisements so as to improve the pushing efficiency of the advertisements.
Referring to fig. 1, a first embodiment of the present application provides a method for pushing an advertisement, which specifically includes the following steps:
s101, acquiring advertisement information of an advertisement to be distributed and user information of a target user.
Optionally, the advertisement information includes at least an advertisement type. The advertisement types are classified according to advertisement content, and specifically, for advertisements provided by banks, the advertisement types can comprise financial product advertisements, interest rate adjustment advertisements, business process adjustment advertisements and the like.
In the expression form of the advertisement, the advertisement can be static words and pictures or a video, wherein when the advertisement is a video, the advertisement information can also comprise the duration of the advertisement.
Optionally, the user information includes at least: user occupation information (for representing the user's current occupation), user family information (for representing the user's family member status), and user preference information (for characterizing the user's personal preferences, e.g., car, sports, reading, etc.).
Optionally, the user information can be obtained by collecting the behavior data analysis of the user on the internet, and the corresponding questionnaire can be pushed to the user for active collection.
S102, processing advertisement information of the advertisement to be released and user information of the target user by using a push time model to obtain the push time of the advertisement to be released, which corresponds to the target user.
The push time model is a neural network model obtained by training a plurality of training samples; each training sample comprises user information of a historical user, advertisement information of a published advertisement which is effectively read by the historical user and actual reading time; the actual reading time refers to the time that the history user effectively reads the published advertisement.
The pushing time of the advertisement to be published corresponding to the target user can be considered as being predicted by using a pushing time model, and the time of the advertisement to be published most likely by the target user or the time of the advertisement to be published most likely by the target user for effective reading.
S103, pushing the advertisement to be released to the target user at the advertisement pushing time to be released corresponding to the target user.
The specific implementation process of step S103 is to detect whether the target user logs in to the designated advertisement publishing platform through any one of the electronic devices in real time after determining the advertisement pushing time to be published corresponding to the target user. For example, if a bank needs to push an advertisement, whether the target user logs in an online bank or an online mall of the bank through a computer or a mobile phone is detected in real time.
After detecting that the target user logs on a designated platform (namely, the target user is online), judging whether the current time is matched with the advertisement pushing time to be distributed of the target user. The matching may be that the absolute value of the difference between the current time and the target user's time to push the advertisement to be sent is less than a certain threshold. For example, it may be set that if the current time is 5 minutes before or after the advertisement push time to be sent by the target user, the two are considered to be matched, that is, if the advertisement push time to be sent by the target user is 20:05, the current time is considered to be matched with the advertisement push time to be sent by the target user as long as the current time is within the time period of 20:00 to 20:10.
If the target user is online and the current time is matched with the to-be-distributed advertisement pushing time of the target user, the to-be-distributed advertisement is pushed to the electronic equipment of the target user which is currently used for logging in the appointed platform.
Pushing the advertisement to be sent to the electronic device means that prompt information of the advertisement to be sent is sent to the electronic device, the electronic device displays the prompt information to a user after receiving the advertisement successfully, the user can select to click or not click the prompt information, and after clicking the prompt information, the electronic device requests a server for a detailed page of the advertisement to be sent, and the obtained detailed page is displayed to the user for the user to read the advertisement.
The application provides an advertisement pushing method, which comprises the steps of obtaining advertisement information of an advertisement to be distributed and user information of a target user; processing advertisement information of advertisements to be released and user information of target users by utilizing a pre-trained push time model to obtain push time of the advertisements to be released, which corresponds to the target users; the push time model is a neural network model obtained by training a plurality of training samples; each training sample comprises user information of a historical user, advertisement information of a published advertisement which is effectively read by the historical user and actual reading time; and pushing the advertisement to be released to the target user at the advertisement pushing time to be released corresponding to the target user.
It can be appreciated that the push time model provided by the scheme is constructed by analyzing the advertisement types of a plurality of users interested in different time periods in the past. The push time model is therefore able to accurately predict when a user with the characteristics described by a particular user information is most likely to be interested in, or is most likely to be effectively reading, what type of advertisement and output the predicted time as push time.
In summary, according to the advertisement pushing scheme provided by the application, the user can effectively read the pushed advertisement with high probability when pushing the advertisement to the user at the predicted pushing time, so that the pushing efficiency of the advertisement is effectively improved.
A second embodiment of the present application provides a training method of a push time model, please refer to fig. 2, the method includes the following steps:
s201, constructing a plurality of training samples by utilizing a plurality of advertisement reading records in an advertisement reading record library.
The advertisement reading record library is a database constructed by an advertisement pushing party (referring to a merchant who needs to push advertisements to users, in particular, the advertisement pushing party may be a bank) by collecting behavior data when the users access a relevant network platform of the advertisement pushing party, and comprises a plurality of advertisement reading records.
An advertisement reading record comprising: a user identifier of a historical user, an advertisement identifier of a published advertisement, and a start time and an end time of the historical user reading the published advertisement.
If an advertisement has been previously pushed to a user, the user is marked as a history user and the advertisement that has been pushed becomes a published advertisement.
The construction process of the advertisement reading record library is described below by taking a bank as an example. The bank can provide web page version of online banking and related online shopping malls on the personal computer, and can provide online banking clients on the mobile phone, and the user can access the network platform to obtain information by using the account after registering the online banking account. For each user, the server of the bank can predict whether the current time is the idle time of the user according to the identity information of the user when the user accesses the network platform. For example, if the user identity is an enterprise employee, 18:00 to 22:00 of weekdays and 7:00 to 22:00 of holidays are predicted to be idle times of the user.
Alternatively, an electronic questionnaire may be pushed to the user, and information filled in by the user in the questionnaire is collected to determine the idle time of the user.
If the current time is the idle time of the user, one or more advertisements are selected from a plurality of advertisements stored in the server to push to the user.
After the server pushes the advertisement to the user, the electronic equipment of the user can pop up related prompt information to prompt the user that the advertisement is pushed currently, and the user can select to click or not click the prompt information. After the user clicks the prompt information, the server pushes the detailed page of the advertisement to the user, the time of the user clicking the prompt information is recorded as the starting time, the user is marked to read the advertisement, and after the user exits the detailed page of the advertisement, the server records the time of the user exiting as the ending time. Thus, the user identifier (which may be a user account number) of the user, the advertisement identifier of the advertisement read by the user (a number may be configured in advance for each advertisement as an advertisement identifier), and the start time and the end time form an advertisement reading record.
Based on the description of the advertisement reading record, the specific method for constructing the training sample by using the advertisement reading record is as follows:
and calculating the difference between the starting time of the advertisement reading record and the ending time of the advertisement reading record aiming at each advertisement reading record to obtain the reading time of the advertisement reading record.
And judging whether the reading time length of each advertisement reading record is longer than the preset effective reading time length or not according to each advertisement reading record.
The effective reading time period is a preset time period, and may be set to 15 seconds, for example. After clicking a prompt message to enter a detailed page of a corresponding advertisement, the user may find that the advertisement is not an interesting advertisement, and then immediately close the detailed page and no longer read the advertisement. By setting the effective reading time length, the reading records of advertisements which are interested by the user can be screened from a plurality of advertisement reading records, and the reading records of advertisements which are not interested by the user are discarded.
And aiming at each advertisement reading record, if the reading time length of the advertisement reading record is longer than the effective reading time length, determining that the advertisement reading record is an effective reading record.
And finally, combining the user information of the historical user of the effective reading record, the advertisement information of the released advertisement of the effective reading record and the starting time of the effective reading record to obtain a training sample corresponding to the effective reading record.
S202, acquiring a preset neural network model, and determining initial model parameters of the neural network model to obtain an initial push model.
The predetermined neural network model may be a neural network model constructed based on any one of the existing neural network frames, which is not limited herein. For example, the predetermined neural network model may be a three-layer Back Propagation (BP) neural network model including an input layer, an output layer and a hidden layer.
After determining the architecture of the neural network model, the number of model parameters involved in the neural network model and the relationships between the model parameters are determined. An initial push model can be obtained only by giving an initial parameter value for each model parameter.
Specifically, the initial model parameters in step S202 may be determined using a genetic algorithm based on acquiring training samples and determining the architecture of the neural network model.
S203, inputting the user information and the advertisement information of each training sample into an initial pushing model to obtain the pushing time of each training sample.
The push time of a training sample is the time predicted by the initial push model that the user involved in the training sample will effectively read the corresponding advertisement (i.e. the reading time period is longer than the effective reading time period, in other words, the user may also consider that the user is interested in the advertisement).
If the training sample includes user information of the user X and advertisement information of the advertisement Y, the push time output by the initial push model is the predicted time when the user X may effectively read the advertisement Y, for example, the output push time is 21:00, which indicates that the initial push model predicts that the user X may effectively read the advertisement Y at 21:00.
S204, calculating model loss of the initial push model according to the push time of each training sample and the actual reading time of the training sample.
Specifically, for each training sample, subtracting the actual reading time from the pushing time of the training sample to obtain a difference value of the training sample, and finally squaring and summing the difference values of all the training samples to obtain the model loss of the initial pushing model.
S205, judging whether the model loss meets a preset model convergence condition.
The model convergence condition may be that the model loss is less than a preset loss threshold. That is, if the current model loss is smaller than the preset loss threshold, the model loss is considered to satisfy the model convergence condition, step S207 is executed, and if the current model loss is greater than or equal to the preset loss threshold, the model loss is considered not to satisfy the model convergence condition, and step S206 is executed.
S206, updating parameters of the initial pushing model according to model loss.
After the step S206 is executed, the initial push model is updated, which is equivalent to updating the initial push model, and an updated initial push model is generated.
After the execution of step S206 is completed, the process returns to step S203.
S207, determining the initial push model as a push time model.
If the current model loss meets the model convergence condition, it is indicated that, for a specific user and a specific type of advertisement, the time predicted by the initial push model for the user to effectively read the advertisement has a high probability that the user actually effectively reads the advertisement, in other words, the initial push model can accurately predict when the specific user is most likely to effectively read the advertisement of the specific type, so that the initial push model at this time can be determined as a push time model and used for predicting the push time of the advertisement.
Optionally, after the push time model is determined, a plurality of verification samples are further constructed by using data in the advertisement reading record library, then the user information and the advertisement information of each verification sample are input into the push time model determined in step S207, the push time model is checked by comparing the push time output by the push time model with the actual reading time of the verification sample, if the check is passed, the prediction is performed by using the push time model, and if the check is not passed, the training is retrained.
Specifically, the process of determining the initial model parameters of the neural network model using the genetic algorithm described in the above step S202 is as follows:
first, a plurality of initial parameter individuals are randomly generated according to the structure of a neural network model.
Each initial parameter individual contains N parameters, wherein the parameters can be real numbers generated randomly, and the value of N depends on the structure of the set neural network model.
After the structure of the neural network model is determined, as described above, how many parameters this model contains is determined. Assuming that the structure of the neural network model determined in step S202 involves 50 parameters, the above N is equal to 50, that is, each initial parameter individual includes 50 real numbers whose values are randomly determined.
And optimizing a plurality of initial parameter individuals by using a genetic algorithm to obtain an optimal parameter individual.
The optimization of the initial parameter individual is specifically as follows:
and determining the genetic probability of each initial parameter individual based on the fitness of the initial parameter individual, wherein the lower the fitness is, the higher the genetic probability is, and simultaneously determining a uniform variation probability. Exchanging every two initial parameter individuals based on the determined genetic probability, namely exchanging a part of parameters of one initial parameter individual with a part of parameters of the other initial parameter individual to obtain two new parameter individuals, and retaining the original initial parameter individuals. On the other hand, each of the initial parameter individuals is mutated based on the mutation probability, that is, the numerical value of a part of the parameters of the initial parameter individuals is randomly adjusted. Through the process, a generation parameter individual can be obtained.
After the first generation parameter individuals are obtained, deleting parameter individuals with overlarge fitness (larger than a certain threshold value) from all parameter individuals (including the initial parameter individuals and the first generation parameter individuals), and repeating the genetic and mutation processes on the rest parameter individuals to obtain the second generation parameter individuals. And similarly, generating third-generation and fourth-generation parameter individuals until parameter individuals with fitness smaller than a certain threshold value or algebraic generation larger than a certain threshold value appear in all parameter individuals. After the algorithm is terminated, the parameter individual with the minimum adaptability in all parameter individuals is the optimal parameter individual.
In this application, the fitness of an individual parameter is defined as the model loss of a neural network model constructed using the individual parameter.
As previously described, each of the initial parameter individuals includes a number of parameters equal to the number of parameters required by the neural network model. After substituting all parameters of an initial parameter individual into the determined neural network model structure for any initial parameter individual, a complete neural network model can be obtained, and then model loss of the neural network model constructed by the parameter individual can be calculated according to the steps S203 and S204,
after generating the optimal parameter individual, each parameter included in the optimal parameter individual is the initial parameter of the neural network model described in step S202.
In combination with the advertisement pushing method described in the foregoing embodiments of the present application, the embodiments of the present application further provide an advertisement pushing device, referring to fig. 3, the device includes the following units:
an obtaining unit 301, configured to obtain advertisement information of an advertisement to be distributed and user information of a target user.
The training unit 302 is configured to train a neural network model by using a plurality of training samples, so as to obtain a push time model.
Each training sample comprises user information of a historical user, advertisement information of a published advertisement which is effectively read by the historical user and actual reading time; the actual reading time refers to the time that the history user effectively reads the published advertisement.
And the processing unit 303 is configured to process the advertisement information of the advertisement to be published and the user information of the target user by using the push time model, so as to obtain the push time of the advertisement to be published, which corresponds to the target user.
And the pushing unit 304 is configured to push the advertisement to be published to the target user at the advertisement to be published pushing time corresponding to the target user.
The training unit 302 trains a neural network model by using a plurality of training samples, and is specifically configured to:
constructing a plurality of training samples by utilizing a plurality of advertisement reading records in an advertisement reading record library;
acquiring a preset neural network model, determining initial model parameters of the neural network model, and obtaining an initial push model;
inputting user information and advertisement information of each training sample into an initial pushing model to obtain pushing time of each training sample;
calculating model loss of an initial push model according to the push time of each training sample and the actual reading time of the training sample;
judging whether the model loss meets a preset model convergence condition or not;
if the model loss does not meet the model convergence condition, updating parameters of an initial pushing model according to the model loss, and returning to execute the input of user information and advertisement information of each training sample into the initial pushing model to obtain the pushing time of each training sample;
and if the model loss meets the model convergence condition, determining the initial push model as a push time model.
Specifically, each advertisement reading record includes: a user identifier of a historical user, an advertisement identifier of a published advertisement, and a starting time and an ending time of the historical user reading the published advertisement;
when the training unit 302 constructs a plurality of training samples by using a plurality of advertisement reading records in the advertisement reading record library, the training unit is specifically configured to:
calculating the difference between the starting time of the advertisement reading record and the ending time of the advertisement reading record aiming at each advertisement reading record to obtain the reading time of the advertisement reading record;
judging whether the reading time length of each advertisement reading record is longer than the preset effective reading time length or not according to each advertisement reading record;
aiming at each advertisement reading record, if the reading time length of the advertisement reading record is longer than the effective reading time length, determining the advertisement reading record as an effective reading record;
and combining the user information of the historical user of the effective reading record and the advertisement information of the released advertisement of the effective reading record with the starting time of the effective reading record to obtain a training sample corresponding to the effective reading record.
Optionally, when the training unit 302 determines the initial model parameters of the neural network model, the training unit is specifically configured to:
randomly generating a plurality of initial parameter individuals according to the structure of the neural network model; wherein each initial parameter individual includes a plurality of parameters;
optimizing a plurality of initial parameter individuals by utilizing a genetic algorithm to obtain an optimal parameter individual;
a plurality of parameters in the optimal parameter individual are determined as initial model parameters of the neural network model.
Optionally, the user information includes at least: user occupation information, user family information and user preference information, and advertisement information at least includes advertisement types.
The specific working principle of the advertisement pushing device provided in this embodiment may refer to the advertisement pushing method provided in any embodiment of the present application, which is not described herein again.
The application provides an advertisement pushing device, wherein an acquisition unit 301 acquires advertisement information of an advertisement to be published and user information of a target user; training unit 302 trains by using a plurality of training samples to obtain a push time model; each training sample comprises user information of a historical user, advertisement information of a published advertisement which is effectively read by the historical user and actual reading time. The processing unit 303 processes advertisement information of the advertisement to be issued and user information of the target user by using a pre-trained push time model to obtain push time of the advertisement to be issued, which corresponds to the target user; the pushing unit 304 pushes the advertisement to be published to the target user at the advertisement to be published pushing time corresponding to the target user.
It can be appreciated that the push time model provided by the scheme is constructed by analyzing the advertisement types of a plurality of users interested in different time periods in the past. The push time model is therefore able to accurately predict when a user with the characteristics described by a particular user information is most likely to be interested in, or is most likely to be effectively reading, what type of advertisement and output the predicted time as push time.
In summary, according to the advertisement pushing scheme provided by the application, the user can effectively read the pushed advertisement with high probability when pushing the advertisement to the user at the predicted pushing time, so that the pushing efficiency of the advertisement is effectively improved.
Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
It should be noted that the terms "first," "second," and the like herein are merely used for distinguishing between different devices, modules, or units and not for limiting the order or interdependence of the functions performed by such devices, modules, or units.
Those skilled in the art can make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (8)
1. A method of advertisement pushing, comprising:
acquiring advertisement information of an advertisement to be published and user information of a target user;
processing the advertisement information of the advertisement to be distributed and the user information of the target user by utilizing a pre-trained push time model to obtain the push time of the advertisement to be distributed, which corresponds to the target user; the push time model is a neural network model obtained by training a plurality of training samples; each training sample comprises user information of a historical user, and the historical user effectively reads advertisement information of a published advertisement and actual reading time; the actual reading time refers to the time for the historical user to effectively read the published advertisement;
pushing the advertisement to be distributed to the target user at the advertisement to be distributed pushing time corresponding to the target user;
the process of training the push time model comprises the following steps:
constructing a plurality of training samples by utilizing a plurality of advertisement reading records in an advertisement reading record library;
acquiring a preset neural network model, and determining initial model parameters of the neural network model to obtain an initial push model;
inputting user information and advertisement information of each training sample into the initial pushing model to obtain pushing time of each training sample;
calculating model loss of the initial push model according to the push time of each training sample and the actual reading time of the training sample;
judging whether the model loss meets a preset model convergence condition or not;
if the model loss does not meet the model convergence condition, updating parameters of the initial pushing model according to the model loss, and returning to execute the step of inputting the user information and the advertisement information of each training sample into the initial pushing model to obtain the pushing time of each training sample;
and if the model loss meets the model convergence condition, determining the initial push model as the push time model.
2. The method of claim 1, wherein each of the advertisement reading records comprises: a user identifier of a historical user and an advertisement identifier of a published advertisement, wherein the historical user reads the starting time and the ending time of the published advertisement;
the constructing a plurality of training samples by using a plurality of advertisement reading records in an advertisement reading record library comprises:
calculating a difference value between the starting time of the advertisement reading record and the ending time of the advertisement reading record aiming at each advertisement reading record to obtain the reading time of the advertisement reading record;
judging whether the reading time length of each advertisement reading record is longer than the preset effective reading time length or not according to each advertisement reading record;
for each advertisement reading record, if the reading time of the advertisement reading record is longer than the effective reading time, determining that the advertisement reading record is an effective reading record;
and combining the user information of the historical user of the effective reading records aiming at each effective reading record, the advertisement information of the released advertisement of the effective reading records and the starting time of the effective reading records to obtain a training sample corresponding to the effective reading records.
3. The method of claim 1, wherein said determining initial model parameters of the neural network model comprises:
randomly generating a plurality of initial parameter individuals according to the structure of the neural network model; wherein each of the initial parameter individuals includes a plurality of parameters;
optimizing the plurality of initial parameter individuals by utilizing a genetic algorithm to obtain an optimal parameter individual;
and determining a plurality of parameters in the optimal parameter individuals as initial model parameters of the neural network model.
4. The method according to claim 1, wherein the user information comprises at least: user occupation information, user family information and user preference information;
the advertisement information includes at least advertisement types.
5. An advertisement pushing apparatus, comprising:
the acquisition unit is used for acquiring advertisement information of the advertisement to be distributed and user information of the target user;
the training unit is used for training a neural network model by utilizing a plurality of training samples to obtain a push time model; each training sample comprises user information of a historical user, and the historical user effectively reads advertisement information of a published advertisement and actual reading time; the actual reading time refers to the time for the historical user to effectively read the published advertisement;
the processing unit is used for processing the advertisement information of the advertisement to be distributed and the user information of the target user by utilizing the pushing time model to obtain the pushing time of the advertisement to be distributed, which corresponds to the target user;
the pushing unit is used for pushing the advertisement to be distributed to the target user at the advertisement to be distributed pushing time corresponding to the target user;
the training unit trains a neural network model by utilizing a plurality of training samples, and is particularly used for when a push time model is obtained:
constructing a plurality of training samples by utilizing a plurality of advertisement reading records in an advertisement reading record library;
acquiring a preset neural network model, and determining initial model parameters of the neural network model to obtain an initial push model;
inputting user information and advertisement information of each training sample into the initial pushing model to obtain pushing time of each training sample;
calculating model loss of the initial push model according to the push time of each training sample and the actual reading time of the training sample;
judging whether the model loss meets a preset model convergence condition or not;
if the model loss does not meet the model convergence condition, updating parameters of the initial pushing model according to the model loss, and returning to execute the step of inputting the user information and the advertisement information of each training sample into the initial pushing model to obtain the pushing time of each training sample;
and if the model loss meets the model convergence condition, determining the initial push model as the push time model.
6. The apparatus of claim 5, wherein each of the advertisement reading records comprises: a user identifier of a historical user and an advertisement identifier of a published advertisement, wherein the historical user reads the starting time and the ending time of the published advertisement;
when the training unit constructs a plurality of training samples by utilizing a plurality of advertisement reading records in the advertisement reading record library, the training unit is specifically used for:
calculating a difference value between the starting time of the advertisement reading record and the ending time of the advertisement reading record aiming at each advertisement reading record to obtain the reading time of the advertisement reading record;
judging whether the reading time length of each advertisement reading record is longer than the preset effective reading time length or not according to each advertisement reading record;
for each advertisement reading record, if the reading time of the advertisement reading record is longer than the effective reading time, determining that the advertisement reading record is an effective reading record;
and combining the user information of the historical user of the effective reading records aiming at each effective reading record, the advertisement information of the released advertisement of the effective reading records and the starting time of the effective reading records to obtain a training sample corresponding to the effective reading records.
7. The apparatus according to claim 5, wherein the training unit is configured to, when determining the initial model parameters of the neural network model:
randomly generating a plurality of initial parameter individuals according to the structure of the neural network model; wherein each of the initial parameter individuals includes a plurality of parameters;
optimizing the plurality of initial parameter individuals by utilizing a genetic algorithm to obtain an optimal parameter individual;
and determining a plurality of parameters in the optimal parameter individuals as initial model parameters of the neural network model.
8. The apparatus of claim 5, wherein the user information comprises at least: user occupation information, user family information and user preference information;
the advertisement information includes at least advertisement types.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010242625.4A CN111461778B (en) | 2020-03-31 | 2020-03-31 | Advertisement pushing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010242625.4A CN111461778B (en) | 2020-03-31 | 2020-03-31 | Advertisement pushing method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111461778A CN111461778A (en) | 2020-07-28 |
| CN111461778B true CN111461778B (en) | 2023-07-14 |
Family
ID=71678455
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010242625.4A Active CN111461778B (en) | 2020-03-31 | 2020-03-31 | Advertisement pushing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111461778B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112399204B (en) * | 2020-11-03 | 2023-01-24 | 重庆农村商业银行股份有限公司 | Alternate type information pushing method, device, equipment and storage medium |
| CN113822696A (en) * | 2021-03-09 | 2021-12-21 | 北京沃东天骏信息技术有限公司 | Information display method, device, system, electronic equipment and medium |
| CN113869964B (en) * | 2021-11-30 | 2022-03-15 | 腾讯科技（深圳）有限公司 | Data processing method and related device |
| CN116228319A (en) * | 2023-05-04 | 2023-06-06 | 广州钛动科技股份有限公司 | Intelligent advertisement putting method and system based on buried points |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105187863A (en) * | 2015-07-31 | 2015-12-23 | 小米科技有限责任公司 | Advertisement playing method and device |
| EP3057055A1 (en) * | 2013-10-08 | 2016-08-17 | Wides Limited | Advertisement informatin sharing system |
| CN107332879A (en) * | 2017-06-05 | 2017-11-07 | 广东欧珀移动通信有限公司 | Method, mobile terminal and storage medium that a kind of information is pushed |
| CN108229994A (en) * | 2016-12-21 | 2018-06-29 | 北京金山云网络技术有限公司 | A kind of information-pushing method and device |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080250450A1 (en) * | 2007-04-06 | 2008-10-09 | Adisn, Inc. | Systems and methods for targeted advertising |
| US20170186043A1 (en) * | 2015-12-29 | 2017-06-29 | Le Holdings (Beijing) Co., Ltd. | Advertisement recommendation method and based on face recogntion and recognition and electronic device |
-
2020
- 2020-03-31 CN CN202010242625.4A patent/CN111461778B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3057055A1 (en) * | 2013-10-08 | 2016-08-17 | Wides Limited | Advertisement informatin sharing system |
| CN105187863A (en) * | 2015-07-31 | 2015-12-23 | 小米科技有限责任公司 | Advertisement playing method and device |
| CN108229994A (en) * | 2016-12-21 | 2018-06-29 | 北京金山云网络技术有限公司 | A kind of information-pushing method and device |
| CN107332879A (en) * | 2017-06-05 | 2017-11-07 | 广东欧珀移动通信有限公司 | Method, mobile terminal and storage medium that a kind of information is pushed |
Non-Patent Citations (1)
| Title |
|---|
| 陆枫 等.基于细粒度标签的在线视频广告投放机制研究.计算机研究与发展.2014,51(12),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111461778A (en) | 2020-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111461778B (en) | Advertisement pushing method and device | |
| US11107118B2 (en) | Management of the display of online ad content consistent with one or more performance objectives for a webpage and/or website | |
| US10614077B2 (en) | Computer system for automated assessment at scale of topic-specific social media impact | |
| US20190104197A1 (en) | Discovering signature of electronic social networks | |
| JP5078910B2 (en) | Estimating advertising quality from observed user behavior | |
| US9245252B2 (en) | Method and system for determining on-line influence in social media | |
| Chen et al. | Predicting the influence of users’ posted information for eWOM advertising in social networks | |
| US8620746B2 (en) | Scoring quality of traffic to network sites | |
| US7594189B1 (en) | Systems and methods for statistically selecting content items to be used in a dynamically-generated display | |
| US20160098735A1 (en) | Marketing channel attribution | |
| US11275748B2 (en) | Influence score of a social media domain | |
| US20120185329A1 (en) | Method and System for Determining Overall Content Values for Content Elements in a Web Network and for Optimizing Internet Traffic Flow Through the Web Network | |
| Jenamani et al. | A stochastic model of e-customer behavior | |
| CN104077711A (en) | Method and system for determining changes in brand awareness after exposure to on-line advertisements | |
| US9858526B2 (en) | Method and system using association rules to form custom lists of cookies | |
| Poulos et al. | Assessing stationarity in web analytics: A study of bounce rates | |
| US20100217668A1 (en) | Optimizing Delivery of Online Advertisements | |
| CN111429161A (en) | Feature extraction method, feature extraction device, storage medium, and electronic apparatus | |
| TWI657395B (en) | Opinion leader related network-based trading system, method, and storage medium | |
| CN113836390B (en) | Resource recommendation method, device, computer equipment and storage medium | |
| CN113450158A (en) | Bank activity information pushing method and device | |
| US20230316106A1 (en) | Method and apparatus for training content recommendation model, device, and storage medium | |
| CN112053245A (en) | Information evaluation method and system | |
| CN117056619A (en) | Method and device for determining user behavior characteristics | |
| WO2019242453A1 (en) | Information processing method and device, storage medium, and electronic 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |