CN109815631A - A kind for the treatment of method and apparatus of game data - Google Patents
A kind for the treatment of method and apparatus of game data Download PDFInfo
- Publication number
- CN109815631A CN109815631A CN201910142466.8A CN201910142466A CN109815631A CN 109815631 A CN109815631 A CN 109815631A CN 201910142466 A CN201910142466 A CN 201910142466A CN 109815631 A CN109815631 A CN 109815631A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- characteristic sequence
- log data
- sequence
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention provides a kind for the treatment of method and apparatus of game data, comprising: obtains games log data；The games log data are screened, target game daily record data is obtained；Characteristic sequence is generated according to the target game daily record data；The characteristic sequence is input to neural network model to be trained, obtains prediction model.This method greatly reduces human cost compared to existing conventional method, improves efficiency.Neural network model is in modeling process simultaneously, take full advantage of high-dimensional historical data, it can plan as a whole local feature and global characteristics, find suitable feature representation, next value is predicted according to the joint probability of preceding value (including previous prediction), accurate prediction can be made to output result.
Description
Technical field
The present invention relates to game technical fields, processing method and a kind of game data more particularly to a kind of game data
Processing unit.
Background technique
In game industry, game user quantity and consumption dynamics be the developing way for influencing game, migration efficiency and
The important evidence of subsequent popularization funds dynamics, and customer churn and reflux are all phenomenons common in game.For leading at present
Stream charges the stronger game on line of dependence to additional content, retain an old user cost cost about obtain one it is new
1/5 spent needed for user, it is also contemplated that a possibility that being lost high consumption user and new person user develop into advanced level user at
This, profit variance will also be increased further.Therefore, accurate reasonable attrition prediction is carried out to user and reflux is predicted, Neng Gouji
When understand and analysis user parade behavior, it is thus understood that the game experiencing of user and the superiority-inferiority of game, to customer churn and return
A possibility that stream, makes a prediction.According to prediction result, operation department is reminded to make reasonable Adjusted Option, keeps this part use
Family enhances game playability, promoting commercial value.
Existing customer churn and the method for reflux specifically include that the simple method manually statisticallyd analyze based on experience and
Choose the method that a small amount of feature and data obtain prediction result by the machine learning on basis；Simple artificial statistics based on experience
The method of analysis, low efficiency, prediction result subjectivity is strong, requires the correlation experience of operator correlation predictive staff high.
Mass users game data is faced simultaneously, and the human cost of the program is high.It chooses a small amount of feature and data and passes through the machine on basis
The method of study obtains prediction result, such as SVM, decision Tree algorithms, by the analysis to these features and differentiates prediction user
A possibility that being lost and flowing back.Weaker in terms of feature representation and model scalability, prediction effect needs to be further improved, and deposits
Accuracy rate it is low, lack high efficiency and the shortcomings that scalability.
Summary of the invention
In view of the above problems, it proposes the embodiment of the present invention and overcomes the above problem or at least partly in order to provide one kind
A kind of processing method of the game data to solve the above problems and a kind of corresponding processing unit of game data.
To solve the above-mentioned problems, the embodiment of the invention discloses a kind of processing methods of game data, comprising:
Obtain games log data；
The games log data are screened, target game daily record data is obtained；
Characteristic sequence is generated according to the target game daily record data；
The characteristic sequence is input to neural network model to be trained, obtains prediction model.
Preferably, described to screen the games log data, obtain target game daily record data the step of include:
The corresponding time shaft of the games log data is constructed sequentially in time；
The games log data for meeting the first preset condition in the time shaft are filtered out by way of vernier, obtain mesh
Mark games log data；
And/or the games log for meeting the second preset condition in the time shaft is filtered out by way of sliding window
Data obtain target game daily record data.
Preferably, the characteristic sequence includes being lost characteristic sequence；It is described to be generated according to the target game daily record data
The step of characteristic sequence includes:
Extract the behavioural characteristic in the target game daily record data；
The loss characteristic sequence is generated according to the behavioural characteristic.
Preferably, the characteristic sequence includes reflux features sequence；It is described to be generated according to the target game daily record data
The step of characteristic sequence includes:
Extract the behavioural characteristic in the target game daily record data；
The reflux features sequence is generated according to the behavioural characteristic.
Preferably, described the characteristic sequence is input to neural network model to be trained, obtain the step of prediction model
Suddenly include:
The loss characteristic sequence is input to neural network model to be trained, obtains attrition prediction model；
And/or be trained the reflux features sequence inputting to neural network model, obtain reflux prediction model.
Preferably, described the loss characteristic sequence is input to neural network model to be trained, obtain attrition prediction
The step of model includes:
Initialize the model parameter of the neural network model；
Loss function is calculated according to the loss characteristic sequence；
Gradient is calculated to loss function derivation and by back-propagation algorithm, by model parameter according to stochastic gradient descent method
It is iterated；
When the frequency of training of model reaches maximum number of iterations or model parameter restrains, the attrition prediction mould is obtained
Type.
Preferably, described to be trained the reflux features sequence inputting to neural network model, obtain reflux prediction
The step of model includes:
Initialize the model parameter of the neural network model；
Loss function is calculated according to the reflux features sequence；
Gradient is calculated to loss function derivation and by back-propagation algorithm, by model parameter according to stochastic gradient descent method
It is iterated；
When the frequency of training of model reaches maximum number of iterations or model parameter restrains, obtains the reflux and predict mould
Type.
Preferably, the method also includes:
It determines the loss characteristic sequence of part or partial reflux features sequence is verifying collection；
Verifying collection is input in corresponding attrition prediction model and/or reflux prediction model, prediction result is obtained；
The attrition prediction model and/or reflux prediction model are verified according to preset evaluation index and prediction result.
The embodiment of the invention also discloses a kind of processing methods of game data, comprising:
Obtain user behavior；
When the user behavior meets trigger condition, the corresponding games log data of the user behavior are obtained；
Characteristic sequence is generated according to the games log data；
The characteristic sequence is input to the prediction model, obtains prediction result.
Preferably, the trigger condition includes the first trigger condition and the second trigger condition；It is described to work as the user behavior
When meeting trigger condition, the step of obtaining the user behavior corresponding games log data, includes:
When the user behavior meets the first trigger condition, the corresponding first games log number of the user behavior is obtained
According to；
Alternatively, obtaining the user behavior corresponding second game day when the user behavior meets the second trigger condition
Will data.
Preferably, described the step of generating characteristic sequence according to the games log data, includes:
It is generated according to the first games log data and is lost characteristic sequence；
And/or reflux features sequence is generated according to the second games log data.
Preferably, the prediction result includes the first prediction result and the second prediction result；It is described by the characteristic sequence
The step of being input to the prediction model, obtaining prediction result include:
The loss characteristic sequence is input to the attrition prediction model, obtains the first prediction result；
And/or by the reflux features sequence inputting to the reflux prediction model, obtain the second prediction result.
The embodiment of the invention also discloses a kind of processing units of game data, comprising:
First daily record data obtains module, for obtaining games log data；
Second log data acquisition module obtains target game log for screening the games log data
Data；
Generation module, for generating characteristic sequence according to the target game daily record data；
Prediction model obtains module, is trained, obtains pre- for the characteristic sequence to be input to neural network model
Survey model.
The embodiment of the invention also discloses a kind of processing units of game data, comprising:
User behavior obtains module, for obtaining user behavior；
Data obtaining module, for obtaining the corresponding trip of the user behavior when the user behavior meets trigger condition
Play daily record data；
Sequence generating module, for generating characteristic sequence according to the games log data；
Prediction result obtains module, for the characteristic sequence to be input to the prediction model, obtains prediction result.
The embodiment of the invention also discloses a kind of electronic equipment, including memory, processor and storage are on a memory simultaneously
The computer program that can be run on a processor, which is characterized in that the processor realizes above-mentioned trip when executing described program
The step of processing for data of playing.
The embodiment of the invention also discloses a kind of computer readable storage medium, deposited on the computer readable storage medium
Computer program is contained, the computer program realizes the processing of above-mentioned game data when being executed by processor the step of.
The embodiment of the present invention includes following advantages:
In the embodiment of the present invention, games log data are obtained；The games log data are screened, target trip is obtained
Play daily record data；Characteristic sequence is generated according to the target game daily record data；The characteristic sequence is input to neural network
Model is trained, and obtains prediction model；This method greatly reduces human cost compared to existing conventional method, improves
Efficiency.Neural network model takes full advantage of high-dimensional historical data, can plan as a whole local feature in modeling process simultaneously
And global characteristics, suitable feature representation is found, next value is predicted according to the joint probability of preceding value (including previous prediction),
Accurate prediction can be made to output result.
Detailed description of the invention
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing
Fig. 1 is a kind of step flow chart of the processing method embodiment one of game data of the embodiment of the present invention；
Fig. 2 is one of embodiment of the present invention characteristic sequence schematic diagram；
Fig. 3 is a kind of step flow chart of the processing method embodiment two of game data of the embodiment of the present invention；
Fig. 4 is a kind of acquisition schematic diagram of target game daily record data of the embodiment of the present invention；
Fig. 5 is a kind of acquisition schematic diagram of target game daily record data of the embodiment of the present invention；
Fig. 6 is a kind of step flow chart of the processing method embodiment three of game data of the embodiment of the present invention；
Fig. 7 is a kind of acquisition schematic diagram of target game daily record data of the embodiment of the present invention；
Fig. 8 is a kind of step flow chart of the processing method example IV of game data of the embodiment of the present invention；
Fig. 9 is a kind of structural block diagram of the processing device embodiment of game data of the embodiment of the present invention
Figure 10 is a kind of structural block diagram of the processing device embodiment of game data of the embodiment of the present invention.
Specific embodiment
The technical issues of in order to keep the embodiment of the present invention solved, technical solution and beneficial effect are more clearly understood, with
The embodiment of the present invention is further described in lower combination accompanying drawings and embodiments.It should be appreciated that specific implementation described herein
Example is only used to explain the present invention, is not intended to limit the present invention.
Referring to Fig.1, a kind of step process of the processing method embodiment one of game data of the embodiment of the present invention is shown
Figure, can specifically include following steps:
Step 101, games log data are obtained；
In the concrete realization, the embodiment of the present invention can be applied in the terminal, for example, mobile phone, tablet computer, individual
Digital assistants, wearable device (such as glasses, wrist-watch) and desktop computer etc..
In embodiments of the present invention, the operating system of mobile terminal may include Android (Android), IOS, Windows
Phone, Windows etc..
In another preferred embodiment of the embodiment of the present invention, the embodiment of the present invention can also be applied in server, should
Server may include server may include device in PC (Personal Computer, PC) service, it is mainframe, small
Type machine, can also include Cloud Server, and the embodiment of the present invention does not limit the type and quantity of server specifically.
Specifically, the number when games log data may include game application operation about game user behavior
According to；The games log data may include all kinds of log fields, timestamp, match information, Transaction Information, game information etc., tool
Body is customer incident type ID (log_id), user's present level (role_grade), customer incident timestamp (timestamp)
Deng, wherein the customer incident type ID may include using technical ability, using article, into copy, leave copy, obtain gold
Money etc., the embodiment of the present invention to this with no restriction.
It is further applicable in the embodiment of the present invention, which can get the game from game server
Daily record data, i.e. the games log data can be stored in one or more game servers, and mobile terminal can pass through net
Network is connect with the game server, gets the games log data by network.
When the embodiment of the present invention is applied to when the server, the server may include game server itself,
The game server can call default process to obtain the games log data for being stored in memory, execute following related
Data handling procedure；It is following to be illustrated by taking mobile terminal as an example.
Step 102, the games log data are screened, obtains target game daily record data；
Further, which can also be screened, obtains target game daily record data；For game
The screening mode of daily record data can be screened according to the preset condition of games log data.
It can be screened according to the time sequencing or registration conditions or online hours condition of games log data；Citing and
Speech, being lost user in predicting problem is the forecasting problem as unit of day, constructs showing for games log data sequentially in time
Form will meet the games log data of the time span of registration conditions as target game daily record data；As from first time
User logs in behavior and starts in continuous 37 days, and there is login behavior in some day again in this 37 days, obtains this day corresponding date
Whether games log data before, logged in using the user in subsequent number of days as label；The i.e. described target game log number
In may include special time period games log data and corresponding label.
Alternatively, reflux user in predicting problem is equally the forecasting problem as unit of day, game is constructed sequentially in time
The time shaft of daily record data, since the user of first time logs in behavior in continuous 90 days, with the cunning of preset duration (such as 52 days)
Dynamic window slides on 90 days time shafts, when detecting there is login behavior, and the 31st to 37 the in sliding window the 30th day
It all without behavior is not logged in when, obtain the user in sliding window first 30 days games log data, by sliding window
Whether user has login behavior to be denoted as corresponding label in 38 days to the 52nd day.
Above-mentioned example is only several citings in the embodiment of the present invention, can also carry out game according to other conditions
The screening of daily record data, the embodiment of the present invention to this with no restriction.
Step 103, characteristic sequence is generated according to the target game daily record data；
It is possible to further generate characteristic sequence according to target game daily record data, specifically, will by Feature Engineering
The target game daily record data generates characteristic sequence, and Feature Engineering refers to the training data for initial data being changed into model
Process, purpose is exactly to obtain better training data feature, so that machine learning model (such as neural network model) approaches this
A upper limit.Feature Engineering can make the performance of model get a promotion.
Specifically, login situation daily in above-mentioned target game daily record data collected, most high can be extracted
Grade, online hours, event major class act frequency, the behavioural characteristic as user's same day；The behavior is used by Feature Engineering
The corresponding characteristic sequence of feature construction.
For example, referring to Fig. 2, one of embodiment of the present invention characteristic sequence schematic diagram is shown, as shown in Fig. 2,
If the time span of target game daily record data is all 30 days, a characteristic sequence is corresponded to every day, such as the 30th day most high
Whether grade the 30th day online hours, logs in, pair of the logarithm of the 30th day 1 frequency of event, the 30th day 2 frequency of event on the 30th day
The 29th day logarithm of the 30th day event n frequency of number, the 29th day highest level event n frequency
Logarithm, the 1st day event n frequency of the 28th day highest level logarithm.
In another embodiment, when the time span of certain target game daily record datas is 30 days inadequate, can pass through
Corresponding Data Position take mend 0 mode, corresponding characteristic sequence is constructed, mainly for what is obtained by way of vernier
Target game daily record data.
Step 104, the characteristic sequence is input to neural network model to be trained, obtains prediction model.
Further, the characteristic sequence is input to neural network model to be trained, obtains prediction model；It needs
Bright, which may include BP network model, depth BP network model, recurrent neural net
Network model, long memory network model and Transformer model etc. in short-term, the embodiment of the present invention to this with no restriction.
It should be noted that the neural network model can unite in such a way that neural network model is come processing feature sequence
Local feature and global characteristics are raised, suitable feature representation is found, in a kind of preferred embodiment, which can
To include Transformer model, used in the Transformer model from attention mechanism (Self-attention
Mechanism), that is, in a preferred embodiment, using based on the neural network model from attention mechanism come processing feature sequence
Column execute attention mechanism inside characteristic sequence, can be considered the implication relation inside search sequence, the spy of this internal relations
Rope and excavation can help to make more accurate prediction to following state.Meanwhile the scalability of the Transformer model
Preferably, convenient in conjunction with surface (numerical value or classification, Time Dependent or sequence rely on).
Specifically, can be directed to by stochastic gradient descent method (SGD, Stochastic Gradient Descent)
Neural network model is trained, and after each characteristic sequence input, model parameter changes according to stochastic gradient descent method
In generation, updates, while exporting training result, finishes when all characteristic sequences input, obtains the prediction model after training.
In the embodiment of the present invention, games log data are obtained；The games log data are screened, target trip is obtained
Play daily record data；Characteristic sequence is generated according to the target game daily record data；The characteristic sequence is input to neural network
Model is trained, and obtains prediction model；This method greatly reduces human cost compared to existing conventional method, improves
Efficiency.Neural network model takes full advantage of high-dimensional historical data, can plan as a whole local feature in modeling process simultaneously
And global characteristics, suitable feature representation is found, next value is predicted according to the joint probability of preceding value (including previous prediction),
Accurate prediction can be made to output result.
Referring to Fig. 3, a kind of step process of the processing method embodiment two of game data of the embodiment of the present invention is shown
Figure, can specifically include following steps:
Step 201, games log data are obtained；
The method of the embodiment of the present invention can solve customer churn forecasting problem, and customer churn forecasting problem can refer to:
There is login behavior for the same day and meet the user of loss testing conditions, predicts whether also to have in following preset time period and step on
Record behavior, if so, being then not to be lost user；It otherwise, is loss user.
Specifically, available first to games log data are obtained, it may include that customer incident type ID, user are current
Grade, customer incident timestamp etc., wherein the customer incident type ID may include using technical ability, using article, into pair
Originally, leave copy, obtain money etc..
Step 202, the corresponding time shaft of the games log data is constructed sequentially in time；
In the embodiment of the present invention, it is contemplated that customer churn forecasting problem is the forecasting problem as unit of day, can be constructed
The corresponding time shaft of games log data constructs the corresponding time shaft of the games log data sequentially in time.
Above-mentioned time shaft is only a kind of specific example of the embodiment of the present invention, can also show form by others
Games log data are showed, show games log data, such as by way of continuous form convenient for games log data
Extract, the embodiment of the present invention to this with no restriction.
Step 203, the games log number for meeting the first preset condition in the time shaft is filtered out by way of vernier
According to acquisition target game daily record data；
Further, the game for meeting the first preset condition in the time shaft can also be filtered out by way of vernier
Daily record data obtains target game daily record data；
Referring to Fig. 4, a kind of acquisition schematic diagram of target game daily record data of the embodiment of the present invention is shown, as shown in figure 4,
It is moved on the corresponding time shaft of games log data in a manner of vernier, acquires corresponding target game daily record data, expanded
Sample size is filled, the richer sample data of content is got.
For example, it was greater than 1 day for contact game less than 37 days, and the new person user that initial grade is 1, passes through Fig. 4
Shown in vernier mode, be sufficient for the data of model training to obtain, can on each time point of time shaft
Obtain the games log data on the same day.Vernier is moved since the user first logs into game along time orientation, once detection
The date intervals for logging in the date Distance Time shaft end of game and current cursor meaning the same day to the user are not less than 7 days,
User games log data whole (including the same day) before the date of vernier meaning are then obtained, and by the user in vernier institute
Login behavior in 7 days after the in a few days phase is denoted as corresponding label and (whether the tag representation customer churn, that is, is not logged in representative stream
Appraxia family), subsequent vernier continues to move to and obtains the qualified games log data of next group.The target game log number
According to may include above-mentioned whole games log data and label.
First preset condition can be moved since the user first logs into game along time orientation for vernier, once
Detect that the user logged in the date intervals of the date Distance Time shaft end of game and current cursor meaning the same day not less than 7
It.
It should be noted that first preset condition be those skilled in the art according to the actual situation and be arranged it is any when
Between condition, registration conditions or combinations thereof condition, the embodiment of the present invention to this with no restriction.
Step 204, the game day for meeting the second preset condition in the time shaft is filtered out by way of sliding window
Will data obtain target game daily record data；
In another preferred embodiment, it can also be filtered out by way of sliding window and be met in the time shaft
The games log data of second preset condition obtain target game daily record data.
Referring to Fig. 5, a kind of acquisition schematic diagram of target game daily record data of the embodiment of the present invention is shown, as shown in figure 5,
It is moved on the corresponding time shaft of games log data in a manner of sliding window, obtains target game daily record data.
For example, 37 days users are not less than for contact game, acquisition can be screened by the way of sliding window
Target game daily record data, as shown in figure 5, the target game daily record data in 37 days can be obtained.In each of time shaft
The games log data on the same day can be obtained on time point.The direction sliding of the sliding window that size is 37 days along the time axis,
When there is login behavior in the 30th day in sliding window, the user the 1st day to the 30th day whole within the scope of sliding window is obtained
Games log data, and login behavior of the 31st day in sliding window into log in the 37th day is denoted as corresponding label
(whether the tag representation customer churn), subsequent sliding window continue to slide and detect the qualified games log number of next group
According to.The target game daily record data may include above-mentioned whole games log data and label.
It should be noted that second preset condition be those skilled in the art according to the actual situation and be arranged it is any when
Between condition, registration conditions or combinations thereof condition, the embodiment of the present invention to this with no restriction.
Step 205, the behavioural characteristic in the target game daily record data is extracted；
It is possible to further extract the behavioural characteristic in the target game daily record data, specifically, can extract
Daily login situation, highest level, online hours, event frequency in above-mentioned target game daily record data collected, as
The behavioural characteristic on user's same day.
Step 206, the loss characteristic sequence is generated according to the behavioural characteristic；
In a kind of specific example of the embodiment of the present invention, the corresponding loss of behavior feature construction is used by Feature Engineering
Characteristic sequence.
Because in the target game daily record data include behavioural characteristic and it is corresponding be not lost user and be lost user
Label, then the loss characteristic sequence be not lost user and be lost that the label of user is same has corresponding relationship.
In view of the otherness between positive and negative sample size, above-mentioned loss characteristic sequence need to be sampled, to obtain
The characteristic sequence of high quality.It is used thus using the loss characteristic sequence down-sampling (subsampling) to non-streaming appraxia family, loss
The method of the loss characteristic sequence up-sampling (upsampling) at family, makes every effort to balance the quantity gap between positive negative sample.
Step 207, the loss characteristic sequence is input to neural network model to be trained, obtains attrition prediction mould
Type.
It is trained it is possible to further which the loss characteristic sequence is input to neural network model, obtains and be lost in advance
Model is surveyed, is being lost after characteristic sequence is input to neural network model, while the prediction result exported, by model parameter
It is constantly adjusted, finally obtains trained attrition prediction model.
It is described that the loss characteristic sequence is input to neural network model in a kind of specific example of the embodiment of the present invention
The step of being trained, obtaining attrition prediction model includes: the model parameter for initializing the neural network model；According to described
It is lost characteristic sequence and calculates loss function (loss function)；It is calculated to loss function derivation and by back-propagation algorithm
Model parameter is iterated by gradient according to stochastic gradient descent method；When the frequency of training of model reach maximum number of iterations or
When person's model parameter restrains, the attrition prediction model is obtained.
Wherein, since the problem of predicting belongs to two classification problems (whether being lost user, or whether reflux user), loss
Function (loss function) can be binary cross entropy (binary cross-entropy),
Specifically, it is 10 that maximum number of iterations in training process (epoch), which can be set, uses and criticize in each iteration
(batch) the mode acceleration model training handled, is set as 256 per a batch of scale.Meanwhile for Optimized model training effect with
Efficiency, stochastic gradient descent algorithm are optimized using Adam optimizer.
The model parameter of random initializtion neural network model first is calculated according to the loss characteristic sequence of tape label and is lost
Then function calculates gradient to loss function derivation and by back-propagation algorithm, model parameter is according to stochastic gradient descent method
It is iterated update；When model training reach maximum number of iterations or model parameter convergence (i.e. front and back twice iterative model join
Number is basicly stable constant), training process terminates, the attrition prediction model after being trained.
In a kind of preferred embodiment of the embodiment of the present invention, the method also includes: determine the loss characteristic sequence of part
For verifying collection；Verifying collection is input in corresponding attrition prediction model, prediction result is obtained；According to preset evaluation index
And prediction result verifies the attrition prediction model.
In the embodiment of the present invention, games log data are obtained；It is corresponding that the games log data are constructed sequentially in time
Time shaft；The games log data for meeting the first preset condition in the time shaft are filtered out by way of vernier, are obtained
Target game daily record data；The game day for meeting the second preset condition in the time shaft is filtered out by way of sliding window
Will data obtain target game daily record data；Extract the behavioural characteristic in the target game daily record data；According to the row
It is characterized and generates the loss characteristic sequence；The loss characteristic sequence is input to neural network model to be trained, is obtained
Attrition prediction model；The embodiment of the present invention solves the problems, such as customer churn prediction, has higher accuracy and preferably may be used
Scalability.
Referring to Fig. 6, a kind of step process of the processing method embodiment three of game data of the embodiment of the present invention is shown
Figure, can specifically include following steps:
Step 301, games log data are obtained；
The method of the embodiment of the present invention can solve user's reflux forecasting problem, and user's reflux forecasting problem can refer to:
For the user that reflux testing conditions are not logged in and met in first time period, predict in next second time period whether
Login behavior is had, if so, being then reflux user；It otherwise, is the user that do not flow back；For example, the first time period can be with
Be 7 days, the second time period can be 15 days, the embodiment of the present invention to this with no restriction.
Specifically, available first to games log data are obtained, it may include that customer incident type ID, user are current
Grade, customer incident timestamp etc., wherein the customer incident type ID may include using technical ability, using article, into pair
Originally, leave copy, obtain money etc..
Step 302, the corresponding time shaft of the games log data is constructed sequentially in time；
In the embodiment of the present invention, it is contemplated that customer churn forecasting problem is the forecasting problem as unit of day, can be constructed
The corresponding time shaft of games log data constructs the corresponding time shaft of the games log data sequentially in time.
Above-mentioned time shaft is only a kind of specific example of the embodiment of the present invention, can also pass through other forms of expression
Show games log data, shows games log data, such as by way of continuous form convenient for games log data
Extract, the embodiment of the present invention to this with no restriction.
Step 303, the game day for meeting the second preset condition in the time shaft is filtered out by way of sliding window
Will data obtain target game daily record data；
Further, it can also be filtered out by way of sliding window and meet the second preset condition in the time shaft
Games log data obtain target game daily record data.
Referring to Fig. 7, a kind of acquisition schematic diagram of target game daily record data of the embodiment of the present invention is shown, as shown in fig. 7,
It is moved on the corresponding time shaft of games log data in a manner of sliding window, obtains target game daily record data.
For example, 52 days users are not less than for contact game, the games log of user is built into one on time
Between sequential deployment time shaft, the games log data on the same day can be obtained on each time point of time shaft.Size is
52 days sliding windows are slided in this time shaft, as shown in fig. 7, having within the 30th day login to go when detecting in sliding window
For, and the 31st to 37 day all without behavior is not logged in when, obtain the user in sliding window first 30 days whole games logs will
In window login behavior in the 38th day to the 52nd day be denoted as it is corresponding reflux prediction label (tag representation user reflux with
It is no, that is, log in and represent reflux user).As sliding window continues to move on a timeline, label and correspondence can be successively obtained
Games log data.I.e. target game daily record data may include above-mentioned whole games log data and label.
Second preset condition can detect there is login behavior, and the 31st to 37 day the 30th day in sliding window to work as
All without being not logged in behavior.
It should be noted that second preset condition be those skilled in the art according to the actual situation and be arranged it is any when
Between condition, registration conditions or combinations thereof condition, the embodiment of the present invention to this with no restriction.
Step 304, the behavioural characteristic in the target game daily record data is extracted；
It is possible to further extract the behavioural characteristic in the target game daily record data, specifically, can extract
Daily login situation, highest level, online hours, event frequency in above-mentioned target game daily record data collected, as
The behavioural characteristic on user's same day.
Step 305, the reflux features sequence is generated according to the behavioural characteristic；
In a kind of specific example of the embodiment of the present invention, the corresponding reflux of behavior feature construction is used by Feature Engineering
Characteristic sequence.
Because in the target game daily record data include behavioural characteristic and it is corresponding be not lost user and reflux user
Label, then the label of user is same has corresponding relationship with do not flow back user and reflux for the reflux features sequence.
In view of the otherness between positive and negative sample size, above-mentioned reflux features sequence need to be sampled, to obtain
The characteristic sequence of high quality.It is used thus using the reflux features sequence down-sampling (subsampling) to the user that do not flow back, reflux
The method of the reflux features sequence up-sampling (upsampling) at family, makes every effort to balance the quantity gap between positive negative sample.
Step 306, the reflux features sequence inputting to neural network model is trained, obtains reflux prediction mould
Type.
It is possible to further which the reflux features sequence inputting to neural network model to be trained, it is pre- to obtain reflux
Model is surveyed, after reflux features sequence inputting to neural network model, while the prediction result exported, by model parameter
It is constantly adjusted, finally obtains trained reflux prediction model.
It is described by the reflux features sequence inputting to neural network model in a kind of specific example of the embodiment of the present invention
The step of being trained, obtaining reflux prediction model includes: the model parameter for initializing the neural network model；According to described
Reflux features sequence calculates loss function；Gradient is calculated to loss function derivation and by back-propagation algorithm, by model parameter
It is iterated according to stochastic gradient descent method；When the frequency of training of model reaches maximum number of iterations or model parameter convergence
When, obtain the reflux prediction model.
In a kind of preferred embodiment of the embodiment of the present invention, the method also includes: determine the reflux features sequence of part
For verifying collection；Verifying collection is input in corresponding reflux prediction model, prediction result is obtained；According to preset evaluation index
And prediction result verifies the reflux prediction model.
Wherein, the preset evaluation index may include model accuracy, recall rate and F1 score etc., the embodiment of the present invention pair
This verifies the reflux prediction model with no restriction, according to above-mentioned specific index and prediction result.
In the embodiment of the present invention, games log data are obtained；It is corresponding that the games log data are constructed sequentially in time
Time shaft；The games log data for meeting the second preset condition in the time shaft are filtered out by way of sliding window,
Obtain target game daily record data；Extract the behavioural characteristic in the target game daily record data；According to the behavioural characteristic
Generate the reflux features sequence；The reflux features sequence inputting to neural network model is trained, it is pre- to obtain reflux
Survey model；The embodiment of the present invention solves the problems, such as that user flows back and predicts there is higher accuracy and better scalability.
Referring to Fig. 8, a kind of step process of the processing method example IV of game data of the embodiment of the present invention is shown
Figure, can specifically include following steps:
Step 401, user behavior is obtained；
In one of embodiment of the present invention core idea, after being trained after prediction model, the prediction can be applied
Model carries out the attrition prediction or reflux prediction of user.
It is possible, firstly, to get user behavior, it can get the user behavior in game process, such as log in behavior
Deng.
Step 402, when the user behavior meets trigger condition, the corresponding games log number of the user behavior is obtained
According to；
Specifically, the trigger condition includes the first trigger condition and the second trigger condition；It is described to work as user's row
When to meet trigger condition, the step of obtaining the user behavior corresponding games log data includes: when the user behavior accords with
When closing the first trigger condition, the corresponding first games log data of the user behavior are obtained；Alternatively, when the user behavior meets
When the second trigger condition, the corresponding second games log data of the user behavior are obtained.
For example, for attrition prediction problem, the first trigger condition may include: to have login behavior and contact trip the same day
Number of days of playing is not less than 30 days users；And/or the same day has login behavior, contact game number of days less than 30 days and initial grade is 1
The user of grade, that is, select the corresponding user group of two kinds of user behaviors.
And for reflux forecasting problem, the second trigger condition may include: to be not logged in and contact game number of days not for continuous 7 days
User less than 37 days selects the corresponding user group of above-mentioned user behavior.Obtain above-mentioned user group corresponding first
Games log data and the second games log data.
Above-mentioned trigger condition is only the citing of the embodiment of the present invention, the trigger condition be those skilled in the art according to
Actual conditions and any condition being arranged, the embodiment of the present invention to this with no restriction.
Step 403, characteristic sequence is generated according to the games log data；
It is further applicable in the embodiment of the present invention, can be generated by Feature Engineering according to the games log data special
Levy sequence；Specific steps are identical as implementation two, the implementation three in the embodiment of the present invention, and details are not described herein.
It is described to generate characteristic sequence according to the games log data in a kind of preferred embodiment of the embodiment of the present invention
Step includes: to be generated to be lost characteristic sequence according to the first games log data；And/or according to second games log
Data generate reflux features sequence.
Step 404, the characteristic sequence is input to the prediction model, obtains prediction result.
It applies in the embodiment of the present invention, characteristic sequence can be input to the prediction model, obtain prediction knot
Fruit, which can calculate each user and be lost probability or reflux probability accordingly, general with the loss probability or reflux
Rate is prediction result.
Specifically, the prediction result includes the first prediction result and the second prediction result；It is described by the feature sequence
The step of column are input to the prediction model, obtain prediction result includes: that the loss characteristic sequence is input to the loss
Prediction model obtains the first prediction result；And/or it by the reflux features sequence inputting to the reflux prediction model, obtains
Second prediction result.
In a specific example, in the potential loss user group for meeting trigger condition, attrition prediction model according to
The loss characteristic sequence of user's A the last 30 days, calculating the customer churn and (no longer logging in) probability at following 7 days is 90%；
Game runs department according to the attrition prediction result (i.e. loss probability) of a large number of users, and discovery is lost the higher user group of probability
(such as acquisition game currency number is few, the viewing cut scene plot time is short), targeted design game scheme improves user and stays
Deposit rate.
In another specific example, in the potential reflux user group for meeting trigger condition, flow back prediction model root
According to first 30 days in user B nearest 37 days reflux features sequences, user reflux (logging at following 15 days online) is calculated
Probability is 85%；Game runs department according to the reflux prediction result (i.e. reflux probability) of user, and discovery reflux probability is higher
User group (stable good friend's quantity is more in such as game, participates in brush copy frequency height of forming a team), design markets or disappears accordingly
Informing mechanism is ceased, new loss user reflux is promoted；For the reflux lower user group of probability, game scheme is adjusted in time, is mentioned
Rise user's reflux ratio.
In the embodiment of the present invention, user behavior is obtained；When the user behavior meets trigger condition, user's row is obtained
For corresponding games log data；Characteristic sequence is generated according to the games log data；The characteristic sequence is input to institute
Prediction model is stated, prediction result is obtained；User information and corresponding attrition prediction result or reflux prediction result are delivered into trip
Play operation department can adjust trip so as to the most possible user group being lost or flow back of game operation department discovery in time
Play scheme promotes game quality, improves user's retention ratio, promotes game commercial value.
It should be noted that for simple description, therefore, it is stated as a series of action groups for embodiment of the method
It closes, but those skilled in the art should understand that, embodiment of that present invention are not limited by the describe sequence of actions, because according to
According to the embodiment of the present invention, some steps may be performed in other sequences or simultaneously.Secondly, those skilled in the art also should
Know, the embodiments described in the specification are all preferred embodiments, and the related movement not necessarily present invention is implemented
Necessary to example.
Referring to Fig. 9, a kind of structural block diagram of the processing device embodiment of game data of the embodiment of the present invention is shown, is had
Body may include following module:
First daily record data obtains module 501, for obtaining games log data；
Second log data acquisition module 502 obtains target game day for screening the games log data
Will data；
Generation module 503, for generating characteristic sequence according to the target game daily record data；
Prediction model obtains module 504, is trained, obtains for the characteristic sequence to be input to neural network model
Prediction model.
Preferably, the second log data acquisition module includes:
Submodule is constructed, for constructing the corresponding time shaft of the games log data sequentially in time；
First obtains submodule, meets the first preset condition in the time shaft for filtering out by way of vernier
Games log data obtain target game daily record data；
And/or second obtain submodule, meet second for being filtered out in the time shaft by way of sliding window
The games log data of preset condition obtain target game daily record data.
The preferably described characteristic sequence includes being lost characteristic sequence；The generation module includes:
First extracting sub-module, for extracting the behavioural characteristic in the target game daily record data；
First generates submodule, for generating the loss characteristic sequence according to the behavioural characteristic.
The preferably described characteristic sequence includes reflux features sequence；The generation module includes:
Second extracting sub-module, for extracting the behavioural characteristic in the target game daily record data；
Second generates submodule, for generating the reflux features sequence according to the behavioural characteristic.
Preferably, the prediction model acquisition module includes:
First training submodule, is trained for the loss characteristic sequence to be input to neural network model, obtains
Attrition prediction model；
And/or second training submodule, for the reflux features sequence inputting to neural network model to be instructed
Practice, obtains reflux prediction model.
Preferably, the first training submodule includes:
First initialization unit, for initializing the model parameter of the neural network model；
First computing unit, for calculating loss function according to the loss characteristic sequence；
First iteration unit, for calculating gradient to loss function derivation and by back-propagation algorithm, by model parameter
It is iterated according to stochastic gradient descent method；
Attrition prediction model obtaining unit reaches maximum number of iterations or model parameter for the frequency of training when model
When convergence, the attrition prediction model is obtained.
Preferably, described to include: by the first training submodule
Second initialization unit, for initializing the model parameter of the neural network model；
Second computing unit, for calculating loss function according to the reflux features sequence；
Secondary iteration unit, for calculating gradient to loss function derivation and by back-propagation algorithm, by model parameter
It is iterated according to stochastic gradient descent method；
Flow back prediction model obtaining unit, reaches maximum number of iterations or model parameter for the frequency of training when model
When convergence, the reflux prediction model is obtained.
Preferably, described device further include:
Determining module, for determining that loss characteristic sequence or the partial reflux features sequence of part are verifying collection；
Prediction result obtains module, for verifying collection to be input to corresponding attrition prediction model and/or reflux in advance
It surveys in model, obtains prediction result；
Authentication module, for verifying the attrition prediction model and/or reflux according to preset evaluation index and prediction result
Prediction model.
Referring to Fig.1 0, a kind of structural block diagram of the processing device embodiment of game data of the embodiment of the present invention is shown,
It can specifically include following module:
User behavior obtains module 601, for obtaining user behavior；
Data obtaining module 602, for it is corresponding to obtain the user behavior when the user behavior meets trigger condition
Games log data；
Sequence generating module 603, for generating characteristic sequence according to the games log data；
Prediction result obtains module 604, for the characteristic sequence to be input to the prediction model, obtains prediction knot
Fruit.
Preferably, the trigger condition includes the first trigger condition and the second trigger condition；The data obtaining module packet
It includes:
First data obtain submodule, for obtaining user's row when the user behavior meets the first trigger condition
For corresponding first games log data；
Alternatively, the second data obtain submodule, for obtaining the use when the user behavior meets the second trigger condition
The corresponding second games log data of family behavior.
Preferably, the sequence generating module includes:
First generates submodule, is lost characteristic sequence for generating according to the first games log data；
And/or second generate submodule, for according to the second games log data generate reflux features sequence.
Preferably, the prediction result includes the first prediction result and the second prediction result；The prediction result obtains mould
Block includes:
First prediction result obtains submodule, for the loss characteristic sequence to be input to the attrition prediction model,
Obtain the first prediction result；
And/or second prediction result obtain submodule, for the reflux features sequence inputting to the reflux to be predicted
Model obtains the second prediction result.
For device embodiment, since it is basically similar to the method embodiment, related so being described relatively simple
Place illustrates referring to the part of embodiment of the method.
The embodiment of the invention also discloses a kind of electronic equipment, including memory, processor and storage are on a memory simultaneously
The computer program that can be run on a processor, which is characterized in that the processor realizes above-mentioned trip when executing described program
The step of processing for data of playing.
The embodiment of the invention also discloses a kind of computer readable storage medium, deposited on the computer readable storage medium
Computer program is contained, the computer program realizes the processing of above-mentioned game data when being executed by processor the step of.
All the embodiments in this specification are described in a progressive manner, the highlights of each of the examples are with
The difference of other embodiments, the same or similar parts between the embodiments can be referred to each other.
It should be understood by those skilled in the art that, the embodiment of the embodiment of the present invention can provide as method, apparatus or calculate
Machine program product.Therefore, the embodiment of the present invention can be used complete hardware embodiment, complete software embodiment or combine software and
The form of the embodiment of hardware aspect.Moreover, the embodiment of the present invention can be used one or more wherein include computer can
With in the computer-usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) of program code
The form of the computer program product of implementation.
The embodiment of the present invention be referring to according to the method for the embodiment of the present invention, terminal device (system) and computer program
The flowchart and/or the block diagram of product describes.It should be understood that flowchart and/or the block diagram can be realized by computer program instructions
In each flow and/or block and flowchart and/or the block diagram in process and/or box combination.It can provide these
Computer program instructions are set to general purpose computer, special purpose computer, Embedded Processor or other programmable data processing terminals
Standby processor is to generate a machine, so that being held by the processor of computer or other programmable data processing terminal devices
Capable instruction generates for realizing in one or more flows of the flowchart and/or one or more blocks of the block diagram
The device of specified function.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing terminal devices
In computer-readable memory operate in a specific manner, so that instruction stored in the computer readable memory generates packet
The manufacture of command device is included, which realizes in one side of one or more flows of the flowchart and/or block diagram
The function of being specified in frame or multiple boxes.
These computer program instructions can also be loaded into computer or other programmable data processing terminal devices, so that
Series of operation steps are executed on computer or other programmable terminal equipments to generate computer implemented processing, thus
The instruction executed on computer or other programmable terminal equipments is provided for realizing in one or more flows of the flowchart
And/or in one or more blocks of the block diagram specify function the step of.
Although the preferred embodiment of the embodiment of the present invention has been described, once a person skilled in the art knows bases
This creative concept, then additional changes and modifications can be made to these embodiments.So the following claims are intended to be interpreted as
Including preferred embodiment and fall into all change and modification of range of embodiment of the invention.
Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning
Covering non-exclusive inclusion, so that process, method, article or terminal device including a series of elements not only wrap
Those elements are included, but also including other elements that are not explicitly listed, or further includes for this process, method, article
Or the element that terminal device is intrinsic.In the absence of more restrictions, being wanted by what sentence "including a ..." limited
Element, it is not excluded that there is also other identical elements in process, method, article or the terminal device for including the element.
Above to a kind of processing method and a kind of processing unit of game data of game data provided by the present invention, into
It has gone and has been discussed in detail, used herein a specific example illustrates the principle and implementation of the invention, the above implementation
The explanation of example is merely used to help understand method and its core concept of the invention；Meanwhile for the general technology people of this field
Member, according to the thought of the present invention, there will be changes in the specific implementation manner and application range, in conclusion this explanation
Book content should not be construed as limiting the invention.
Claims (16)
1. a kind of processing method of game data characterized by comprising
Obtain games log data；
The games log data are screened, target game daily record data is obtained；
Characteristic sequence is generated according to the target game daily record data；
The characteristic sequence is input to neural network model to be trained, obtains prediction model.
2. being obtained the method according to claim 1, wherein described screen the games log data
The step of target game daily record data includes:
The corresponding time shaft of the games log data is constructed sequentially in time；
The games log data for meeting the first preset condition in the time shaft are filtered out by way of vernier, obtain target trip
Play daily record data；
And/or the games log data for meeting the second preset condition in the time shaft are filtered out by way of sliding window,
Obtain target game daily record data.
3. according to the method described in claim 2, it is characterized in that, the characteristic sequence includes being lost characteristic sequence；Described
Include: according to the step of target game daily record data generation characteristic sequence
Extract the behavioural characteristic in the target game daily record data；
The loss characteristic sequence is generated according to the behavioural characteristic.
4. according to the method described in claim 2, it is characterized in that, the characteristic sequence includes reflux features sequence；Described
Include: according to the step of target game daily record data generation characteristic sequence
Extract the behavioural characteristic in the target game daily record data；
The reflux features sequence is generated according to the behavioural characteristic.
5. the method according to claim 3 or 4, which is characterized in that described that the characteristic sequence is input to neural network
The step of model is trained, and obtains prediction model include:
The loss characteristic sequence is input to neural network model to be trained, obtains attrition prediction model；
And/or be trained the reflux features sequence inputting to neural network model, obtain reflux prediction model.
6. the method according to claim 3 or 4, which is characterized in that described that the loss characteristic sequence is input to nerve
Network model is trained, obtain attrition prediction model the step of include:
Initialize the model parameter of the neural network model；
Loss function is calculated according to the loss characteristic sequence；
Gradient is calculated to loss function derivation and by back-propagation algorithm, model parameter is carried out according to stochastic gradient descent method
Iteration；
When the frequency of training of model reaches maximum number of iterations or model parameter restrains, the attrition prediction model is obtained.
7. the method according to claim 3 or 4, which is characterized in that described by the reflux features sequence inputting to nerve
Network model is trained, and is obtained the step of flowing back prediction model and is included:
Initialize the model parameter of the neural network model；
Loss function is calculated according to the reflux features sequence；
Gradient is calculated to loss function derivation and by back-propagation algorithm, model parameter is carried out according to stochastic gradient descent method
Iteration；
When the frequency of training of model reaches maximum number of iterations or model parameter restrains, the reflux prediction model is obtained.
8. the method according to claim 3 or 4, which is characterized in that the method also includes:
It determines the loss characteristic sequence of part or partial reflux features sequence is verifying collection；
Verifying collection is input in corresponding attrition prediction model and/or reflux prediction model, prediction result is obtained；
The attrition prediction model and/or reflux prediction model are verified according to preset evaluation index and prediction result.
9. a kind of processing method of game data characterized by comprising
Obtain user behavior；
When the user behavior meets trigger condition, the corresponding games log data of the user behavior are obtained；
Characteristic sequence is generated according to the games log data；
The characteristic sequence is input to the prediction model, obtains prediction result.
10. according to the method described in claim 9, it is characterized in that, the trigger condition includes the first trigger condition and second
Trigger condition；It is described when the user behavior meets trigger condition, obtain the corresponding games log data of the user behavior
Step includes:
When the user behavior meets the first trigger condition, the corresponding first games log data of the user behavior are obtained；
Alternatively, obtaining the corresponding second games log number of the user behavior when the user behavior meets the second trigger condition
According to.
11. according to the method described in claim 10, it is characterized in that, described generate feature sequence according to the games log data
The step of column includes:
It is generated according to the first games log data and is lost characteristic sequence；
And/or reflux features sequence is generated according to the second games log data.
12. according to the method for claim 11, which is characterized in that the prediction result includes the first prediction result and second
Prediction result；Described that the characteristic sequence is input to the prediction model, the step of obtaining prediction result, includes:
The loss characteristic sequence is input to the attrition prediction model, obtains the first prediction result；
And/or by the reflux features sequence inputting to the reflux prediction model, obtain the second prediction result.
13. a kind of processing unit of game data characterized by comprising
First daily record data obtains module, for obtaining games log data；
Second log data acquisition module obtains target game daily record data for screening the games log data；
Generation module, for generating characteristic sequence according to the target game daily record data；
Prediction model obtains module, is trained for the characteristic sequence to be input to neural network model, obtains prediction mould
Type.
14. a kind of processing unit of game data characterized by comprising
User behavior obtains module, for obtaining user behavior；
Data obtaining module, for when the user behavior meets trigger condition, obtaining the user behavior corresponding game day
Will data；
Sequence generating module, for generating characteristic sequence according to the games log data；
Prediction result obtains module, for the characteristic sequence to be input to the prediction model, obtains prediction result.
15. a kind of electronic equipment including memory, processor and stores the calculating that can be run on a memory and on a processor
Machine program, which is characterized in that the processor realizes the trip as described in any one of claims 1 to 12 when executing described program
The step of processing for data of playing.
16. a kind of computer readable storage medium, which is characterized in that be stored with computer on the computer readable storage medium
Program, the game data as described in any one of claims 1 to 12 is realized when the computer program is executed by processor
The step of processing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910142466.8A CN109815631A (en) | 2019-02-26 | 2019-02-26 | A kind for the treatment of method and apparatus of game data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910142466.8A CN109815631A (en) | 2019-02-26 | 2019-02-26 | A kind for the treatment of method and apparatus of game data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109815631A true CN109815631A (en) | 2019-05-28 |
Family
ID=66607443
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910142466.8A Pending CN109815631A (en) | 2019-02-26 | 2019-02-26 | A kind for the treatment of method and apparatus of game data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109815631A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111507471A (en) * | 2020-03-03 | 2020-08-07 | 上海喜马拉雅科技有限公司 | Model training method, device, equipment and storage medium |
| CN111821694A (en) * | 2020-07-24 | 2020-10-27 | 北京达佳互联信息技术有限公司 | Loss prevention method and device for new game user, electronic equipment and storage medium |
| CN111861588A (en) * | 2020-08-06 | 2020-10-30 | 网易（杭州）网络有限公司 | Training method of loss prediction model, player loss reason analysis method and player loss reason analysis device |
| CN112365290A (en) * | 2020-11-26 | 2021-02-12 | 上海触乐信息科技有限公司 | Method, device, storage medium and server for improving user retention rate of application |
| CN112843726A (en) * | 2021-03-15 | 2021-05-28 | 网易（杭州）网络有限公司 | Intelligent agent processing method and device |
| CN113082711A (en) * | 2021-03-22 | 2021-07-09 | 北京达佳互联信息技术有限公司 | Game robot control method, game robot control device, server, and storage medium |
| CN113191810A (en) * | 2021-04-30 | 2021-07-30 | 网易（杭州）网络有限公司 | Game index prediction method and device and electronic equipment |
| CN113869501A (en) * | 2021-10-19 | 2021-12-31 | 京东科技信息技术有限公司 | Neural network generation method and device, electronic equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107507036A (en) * | 2017-08-28 | 2017-12-22 | 深圳市诚壹科技有限公司 | The method and terminal of a kind of data prediction |
| CN107609708A (en) * | 2017-09-25 | 2018-01-19 | 广州赫炎大数据科技有限公司 | A kind of customer loss Forecasting Methodology and system based on mobile phone games shop |
-
2019
- 2019-02-26 CN CN201910142466.8A patent/CN109815631A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107507036A (en) * | 2017-08-28 | 2017-12-22 | 深圳市诚壹科技有限公司 | The method and terminal of a kind of data prediction |
| CN107609708A (en) * | 2017-09-25 | 2018-01-19 | 广州赫炎大数据科技有限公司 | A kind of customer loss Forecasting Methodology and system based on mobile phone games shop |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111507471A (en) * | 2020-03-03 | 2020-08-07 | 上海喜马拉雅科技有限公司 | Model training method, device, equipment and storage medium |
| CN111507471B (en) * | 2020-03-03 | 2023-11-17 | 上海喜马拉雅科技有限公司 | Model training method, device, equipment and storage medium |
| CN111821694A (en) * | 2020-07-24 | 2020-10-27 | 北京达佳互联信息技术有限公司 | Loss prevention method and device for new game user, electronic equipment and storage medium |
| CN111821694B (en) * | 2020-07-24 | 2024-05-21 | 北京达佳互联信息技术有限公司 | Loss prevention method and device for new game user, electronic equipment and storage medium |
| CN111861588A (en) * | 2020-08-06 | 2020-10-30 | 网易（杭州）网络有限公司 | Training method of loss prediction model, player loss reason analysis method and player loss reason analysis device |
| CN111861588B (en) * | 2020-08-06 | 2023-10-31 | 网易（杭州）网络有限公司 | Training method of loss prediction model, player loss reason analysis method and player loss reason analysis device |
| CN112365290A (en) * | 2020-11-26 | 2021-02-12 | 上海触乐信息科技有限公司 | Method, device, storage medium and server for improving user retention rate of application |
| CN112843726A (en) * | 2021-03-15 | 2021-05-28 | 网易（杭州）网络有限公司 | Intelligent agent processing method and device |
| CN113082711A (en) * | 2021-03-22 | 2021-07-09 | 北京达佳互联信息技术有限公司 | Game robot control method, game robot control device, server, and storage medium |
| CN113082711B (en) * | 2021-03-22 | 2023-08-29 | 北京达佳互联信息技术有限公司 | Game robot control method, game robot control device, server and storage medium |
| CN113191810A (en) * | 2021-04-30 | 2021-07-30 | 网易（杭州）网络有限公司 | Game index prediction method and device and electronic equipment |
| CN113869501A (en) * | 2021-10-19 | 2021-12-31 | 京东科技信息技术有限公司 | Neural network generation method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109815631A (en) | A kind for the treatment of method and apparatus of game data | |
| CN111784348B (en) | Account risk identification method and device | |
| CN106886915B (en) | Advertisement click estimation method based on time attenuation sampling | |
| US10831827B2 (en) | Automatic extraction of user mobility behaviors and interaction preferences using spatio-temporal data | |
| CN109597844B (en) | Core user mining method and system based on deep neural network and graph network | |
| CN110503531A (en) | The dynamic social activity scene recommended method of timing perception | |
| CN111177473B (en) | Personnel relationship analysis method, device and readable storage medium | |
| CN109784997B (en) | Short video active user prediction method based on big data | |
| CN110427560A (en) | A kind of model training method and relevant apparatus applied to recommender system | |
| CN109767269A (en) | A kind for the treatment of method and apparatus of game data | |
| CN109960761A (en) | Information recommendation method, device, equipment and computer readable storage medium | |
| CN107181776A (en) | A kind of data processing method and relevant device, system | |
| CN108734327A (en) | A kind of data processing method, device and server | |
| Hanga et al. | A graph-based approach to interpreting recurrent neural networks in process mining | |
| CN108846695A (en) | The prediction technique and device of terminal replacement cycle | |
| CN109754290A (en) | A kind for the treatment of method and apparatus of game data | |
| KR102223640B1 (en) | Cloud-based personalized contents subscription service providing system and method thereof | |
| CN111408143A (en) | Game payment prediction method, model training method and device | |
| CN114721898A (en) | Edge cloud server utilization rate prediction method and device based on boosting algorithm and storage medium | |
| CN109978575A (en) | A kind of method and device excavated customer flow and manage scene | |
| CN108345857A (en) | A kind of region crowd density prediction technique and device based on deep learning | |
| CN114238764A (en) | Course recommendation method, device and equipment based on recurrent neural network | |
| Yi et al. | Purchase behavior prediction in m-commerce with an optimized sampling methods | |
| CN112245934B (en) | Data analysis method, device and equipment for virtual resources in virtual scene application | |
| CN111325255B (en) | Specific crowd delineating method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |