CN107153887A - A kind of mobile subscriber's behavior prediction method based on convolutional neural networks - Google Patents
A kind of mobile subscriber's behavior prediction method based on convolutional neural networks Download PDFInfo
- Publication number
- CN107153887A CN107153887A CN201710243498.8A CN201710243498A CN107153887A CN 107153887 A CN107153887 A CN 107153887A CN 201710243498 A CN201710243498 A CN 201710243498A CN 107153887 A CN107153887 A CN 107153887A
- Authority
- CN
- China
- Prior art keywords
- behavior
- mobile subscriber
- historical
- convolution
- record
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06Q30/0242—Determining effectiveness of advertisements
-
- 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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
-
- 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
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M1/00—Substation equipment, e.g. for use by subscribers
- H04M1/72—Mobile telephones; Cordless telephones, i.e. devices for establishing wireless links to base stations without route selection
- H04M1/724—User interfaces specially adapted for cordless or mobile telephones
- H04M1/72448—User interfaces specially adapted for cordless or mobile telephones with means for adapting the functionality of the device according to specific conditions
- H04M1/72457—User interfaces specially adapted for cordless or mobile telephones with means for adapting the functionality of the device according to specific conditions according to geographic location
Abstract
The invention discloses a kind of mobile subscriber's behavior prediction method based on convolutional neural networks, step includes：Characteristic processing is carried out to mobile subscriber's historical data, convolution user is selected according to historical record number, the two-dimentional historical data unit of convolution user is built, convolutional neural networks behavior prediction model is trained；Judge that targeted customer predicts targeted customer for convolution, build two dimension target user data cell input convolutional neural networks behavior prediction model, prediction obtains the probability of each behavior label of targeted customer.The present invention can effectively improve the accuracy rate of mobile subscriber's behavior prediction, while reducing the workload of mobile subscriber's behavioral data Feature Engineering.
Description
Technical field
The invention belongs to mobile personalized service field, more particularly to a kind of mobile subscriber's row based on convolutional neural networks
For Forecasting Methodology.
Background technology
Deep neural network is a kind of neutral net for possessing at least one hidden layer, compared to shallow-layer neutral net, is had more
Level higher abstraction hierarchy is provided for model, thus improve the ability of model.Convolutional neural networks are by one or many
Individual convolutional layer and the full-mesh on top layer composition, while also including associated weights and pond layer, this structure causes convolutional Neural
Network can utilize the two-dimensional structure of input data.Compare other depth, feedforward neural network, and convolutional neural networks need to estimate
The parameter of meter is less, makes a kind of deep learning structure for having much attraction.
With the fast development of mobile Internet, daily life is increasingly dependent on mobile Internet with work,
How user's behavior prediction is moved using growing mobile subscriber's big data, have become mobile personalized service neck
Urgent problem to be solved in terms of domain, such as mobile e-business, mobile location-based service, the prediction of moving advertising clicking rate.
The content of the invention
It is an object of the invention to the shortcoming and deficiency for overcoming prior art, there is provided a kind of shifting based on convolutional neural networks
Dynamic user's behavior prediction method, the accuracy to improve mobile subscriber's behavior prediction.
The purpose of the present invention is realized by following technical scheme：A kind of mobile subscriber's behavior based on convolutional neural networks
Forecasting Methodology, comprises the following steps：
S1, to mobile subscriber's historical data carry out characteristic processing；Mobile subscriber's historical data includes mobile subscriber's client
Information, mobile subscriber's network link information, mobile subscriber's characteristic information and mobile subscriber's historical behavior information；
S2, determine history sliding window size N, mobile subscriber's historical behavior record number is more than in selection mobile subscriber's historical data
Or the mobile subscriber equal to N is used as convolution user；
Mobile subscriber's historical behavior record refers to that behavioral data of the mobile subscriber at some moment is recorded；One movement
A behavior act, behavior act information and the behavior act of user's history behavior record including mobile subscriber occur when
Between, object, in addition to mobile subscriber's client-side information during behavior act generation, mobile subscriber's network link information and mobile use
Family characteristic information；
S3, mobile subscriber's historical behavior to each convolution user are recorded, and the time order and function occurred by behavior act is arranged,
Constitute the behavior record list of each convolution user；
Then the two-dimentional historical data unit of convolution user is built：
In the behavior record list of each convolution user, the sliding window for being N with length slides selection successively, each to obtain
To the convolved data group recorded comprising N bar mobile subscribers historical behavior, by N bar mobile subscriber's historical behaviors in convolved data group
The two-dimentional historical data unit that record concatenation is represented into two-dimensional matrix；
In wherein two-dimentional historical data unit, mobile subscriber's historical behavior that the time of behavior act generation is newest is remembered
The behavior act information of record, as the behavior label of the behavior label field of two-dimentional historical data unit, while, it is necessary to two dimension is gone through
The action message of time that behavior act in history data cell occurs newest mobile subscriber's historical behavior record is emptied,
Fill default value 0；
S4, two-dimentional historical data unit and its behavior label by all convolution users, are used as input data, input to volume
Behavior prediction model training is carried out in product neutral net, the parameter of convolutional neural networks behavior prediction model is obtained, and then obtained
Convolutional neural networks behavior prediction model；
S5, the mobile subscriber's historical behavior record number for obtaining targeted customer, mobile subscriber's historical behavior note of targeted customer
Number is recorded if greater than or equal to N-1, then targeted customer is that convolution predicts targeted customer, otherwise targeted customer is other prediction targets
User；
The time that the current request time of S6, acquisition convolution prediction targeted customer occurs as behavior act, obtain convolution
Targeted customer current mobile subscriber's client-side information, mobile subscriber's network link information and mobile subscriber's characteristic information are predicted,
Mobile subscriber's historical behavior record of the current convolution prediction targeted customer to be predicted of generation one, this record, which does not include, currently will
The behavior label of prediction；
Targeted customer is predicted convolution, and housing choice behavior acts the time occurred nearest N-1 bar mobile subscriber's historical behaviors
Record, and current mobile subscriber's historical behavior record to be predicted, in chronological sequence, composition includes N bar mobile subscriber's history
The two dimension target user data cell of behavior record；The nearest N-1 bar mobile subscribers of wherein selected action time of origin go through
The Records of the Historian records the behavior label for containing respective record time point；
S7, by above-mentioned two dimension target user data cell input convolutional neural networks behavior prediction model in, obtain convolution
Prediction targeted customer takes the probability of each behavior label.
It is preferred that, characteristic processing is carried out to mobile subscriber's historical data in step S1, including data cleansing, missing values are filled out
Fill, quantize and data normalized；Quantize and refer to that carrying out Hash coding to character or character string type data obtains whole
Several identifiers.
It is preferred that, history sliding window size N is set as a fixed value or determined by algorithm in step S2.
It is preferred that, history sliding window size N values are 10 or take mobile subscriber's historical behavior of all mobile subscribers to record
Two several quantiles.
It is preferred that, slide and refer in the behavior record list of convolution user successively in step S3, sliding window is by backward
Increase by one before last mobile subscriber's historical behavior record grand window, sliding window in the preceding step of movement 1, sliding window
Individual new mobile subscriber's historical behavior record, wherein sliding window moves forward 1 step, until the behavior record of traversal convolution user
List.
It is preferred that, behavior prediction model training is carried out in step S4 and refers to carry out convolutional Neural net using back-propagation algorithm
Network model training.
It is preferred that, convolutional neural networks behavior prediction model includes convolutional layer, pond layer, optional regularization in step S4
Layer, intensive full Connection Neural Network layer, SoftMax output layers；Specifically, convolutional neural networks behavior prediction model is including defeated
Enter layer, the first convolutional layer, Batch_normalize layers, the second convolutional layer, pooling layers of the first Max, the first Dropout layers,
3rd convolutional layer, pooling layers of the 2nd Max, the 2nd Dropout layers and output layer；The convolution kernel size of wherein the first convolutional layer
It is 20 for 5x3, convolution kernel number, the convolution kernel size of the second convolutional layer and the 3rd convolutional layer is that 3x3, convolution kernel number are 15；
First Dropout layers and the 2nd Dropout layers as restraint layer, the number come with 0.5 probability dropping preceding layer network delivery
It is worth result；The pooling layers of max using 2x2 sizes of Max are operated.
Further, the convolutional layer number wherein in convolutional neural networks behavior prediction model is empirically determined or passes through
The unit for possessing direct-connected path is introduced, is semi-automatedly determined.
It is preferred that, convolutional neural networks behavior prediction model is output as multiple probability to each two-dimentional historical data unit
Value, represents that the behavior label field of the two-dimentional historical data unit obtains the probability of each behavior label.
It is preferred that, to other predictions targeted customer, predict that other predictions targeted customer takes respectively using the second forecast model
The probability of individual behavior label；Second forecast model refers to one kind in popularity model, linear prediction model, or popularity model
With the combination of linear prediction model.
The present invention compared with prior art, has the following advantages that and beneficial effect：
1. the present invention improves the accuracy of mobile subscriber's behavior prediction.
2. the present invention realizes user's behavior prediction using convolutional neural networks, the feature work of training forecast model is reduced
Journey workload, adapts to different complicated movement network application environments.
Brief description of the drawings
Fig. 1 is the step flow chart of mobile subscriber's behavior prediction method based on convolutional neural networks in embodiment.
Fig. 2 is the comparison diagram of the effect of the model and linear regression model (LRM) proposed using embodiment.
Embodiment
With reference to embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are not limited
In this.
A kind of mobile subscriber's behavior prediction method based on convolutional neural networks, as shown in figure 1, comprising the following steps that：
1st step, to mobile subscriber's historical data carry out characteristic processing, including data cleansing, Missing Data Filling, quantize and
Data normalization processing；
Mobile subscriber's historical data includes mobile subscriber's client-side information, mobile subscriber's network link information, mobile subscriber
Characteristic information and mobile subscriber's historical behavior information.
Above-mentioned quantize refers to carry out character or character string type data the identifier that Hash coding obtains integer.
2nd step, determines history sliding window size N；Select mobile subscriber's historical behavior record number in mobile subscriber's historical data
Mobile subscriber more than or equal to N is used as convolution user；Wherein N values are 10.
The length N of wherein sliding window can be one fixed value of setting or be determined by algorithm, such as selection institute
The mobile subscriber's historical behavior for having mobile subscriber records two quantiles of number.
Mobile subscriber's historical behavior record refers to that behavioral data of the mobile subscriber at some moment is recorded；One movement
A behavior act, behavior act information and the behavior act of user's history behavior record including mobile subscriber occur when
Between, object etc., in addition to mobile subscriber's client-side information, mobile subscriber's network link information and movement during behavior act generation
User's characteristic information.
3rd step, is recorded to mobile subscriber's historical behavior of each convolution user, and the time order and function occurred by behavior act is arranged
Row, constitute the behavior record list of each convolution user.
In the behavior record list of each convolution user, the sliding window for being N with length slides selection successively, each to obtain
To the convolved data group recorded comprising N bar mobile subscribers historical behavior, by N bar mobile subscriber's historical behaviors in convolved data group
The two-dimentional historical data unit that record concatenation is represented into two-dimensional matrix.
In wherein two-dimentional historical data unit, mobile subscriber's historical behavior that the time of behavior act generation is newest is remembered
The behavior act information of record, as the behavior label of the behavior label field of two-dimentional historical data unit, while, it is necessary to two dimension is gone through
The action message of time that behavior act in history data cell occurs newest mobile subscriber's historical behavior record is emptied,
Fill default value 0.Behavior label refers to behavior act information, such as clicks on, and downloads, browses, behavior act information is needs
The label of prediction.
Slide and refer in the behavior record list of convolution user successively, sliding window moves 1 step, sliding window from the front to the back
Increase new mobile subscriber's history before intraoral last mobile subscriber's historical behavior record grand window, sliding window
Behavior record, wherein sliding window move forward 1 step, until the behavior record list of traversal convolution user.
4th step, by the two-dimentional historical data unit and its behavior label of all convolution users, is used as input data, input
Behavior prediction model training is carried out into convolutional neural networks, the parameter of convolutional neural networks behavior prediction model is obtained, and then
Obtain convolutional neural networks behavior prediction model.
Behavior prediction model training is carried out to refer to carry out convolutional neural networks model training using back-propagation algorithm.
Convolutional neural networks behavior prediction model includes convolutional layer, pond layer, optional regularization layer, intensive full connection
Neural net layer, SoftMax output layers.Specifically, convolutional neural networks behavior prediction model includes input layer, the first convolution
Layer, Batch_normalize layers, the second convolutional layer, pooling layers of the first Max, the first Dropout layers, the 3rd convolutional layer, the
Two Maxpooling layers, the 2nd Dropout layers and output layer.The convolution kernel size of wherein the first convolutional layer is 5x3, convolution check figure
Mesh is 20, and the convolution kernel size of the second convolutional layer and the 3rd convolutional layer is that 3x3, convolution kernel number are 15；First Dropout layers
With the 2nd Dropout layers as restraint layer, the numerical result come with 0.5 probability dropping preceding layer network delivery；Max
The pooling layers of max using 2x2 sizes is operated.
The convolution number of plies wherein in convolutional neural networks behavior prediction model can be empirically determined, can also be gathered around by introducing
There is the unit of direct-connected path, semi-automatedly determine.
Convolutional neural networks behavior prediction model is output as multiple probable values to each two-dimentional historical data unit, represents
The behavior label field of the two-dimentional historical data unit obtains the probability of each behavior label.
5th step, obtains mobile subscriber's historical behavior record number of targeted customer, mobile subscriber's historical behavior of targeted customer
Number is recorded if greater than or equal to N-1, then targeted customer is that convolution predicts targeted customer, otherwise targeted customer is other prediction mesh
Mark user.
To other predictions targeted customer, predict that other predictions targeted customer takes each behavior mark using the second forecast model
The probability of label.Second forecast model refers to one kind in popularity model, linear prediction model (such as linear regression model (LRM)), or stream
The combination of row degree model and linear prediction model.
6th step, obtains the time that the current request time of convolution prediction targeted customer occurs as behavior act, is rolled up
The current mobile subscriber's client-side information of product prediction targeted customer, mobile subscriber's network link information and mobile subscriber's feature letter
Breath, generates mobile subscriber's historical behavior record of a current convolution prediction targeted customer to be predicted, this record, which does not include, works as
Before the behavior label to be predicted.
Targeted customer is predicted convolution, and housing choice behavior acts the time occurred nearest N-1 bar mobile subscriber's historical behaviors
Record, and current mobile subscriber's historical behavior record to be predicted, in chronological sequence, composition includes N bar mobile subscriber's history
The two dimension target user data cell of behavior record.The nearest N-1 bar mobile subscribers of wherein selected action time of origin go through
The Records of the Historian records the behavior label for containing respective record time point.
7th step, above-mentioned two dimension target user data cell is inputted in convolutional neural networks behavior prediction model, rolled up
Product prediction targeted customer takes the probability of each behavior label.
Fig. 2 is the result of the model on True Data, proposed using the present embodiment and the effect of linear regression model (LRM)
Comparison diagram, wherein transverse axis are the time of data.Repeating 5 times and testing to take the value of average result.The side that the present embodiment is proposed
Method improves the accuracy of mobile subscriber's behavior prediction.General linear regression model (LRM), has AUC relative in full dataset
2% lifting；If only considering the user with enough historical information bar numbers, the model opposite linear regression model of the present embodiment
AUC lifting can reach 4%.
Above-described embodiment is preferably embodiment, but embodiments of the present invention are not by above-described embodiment of the invention
Limitation, other any Spirit Essences without departing from the present invention and the change made under principle, modification, replacement, combine, simplification,
Equivalent substitute mode is should be, is included within protection scope of the present invention.
Claims (10)
1. a kind of mobile subscriber's behavior prediction method based on convolutional neural networks, it is characterised in that comprise the following steps：
S1, to mobile subscriber's historical data carry out characteristic processing；Mobile subscriber's historical data include mobile subscriber's client-side information,
Mobile subscriber's network link information, mobile subscriber's characteristic information and mobile subscriber's historical behavior information；
S2, determine history sliding window size N, mobile subscriber's historical behavior record number is more than or waited in selection mobile subscriber's historical data
Convolution user is used as in N mobile subscriber；
Mobile subscriber's historical behavior record refers to that behavioral data of the mobile subscriber at some moment is recorded；One mobile subscriber
Historical behavior record includes the behavior act of mobile subscriber, the time of behavior act information and behavior act generation, right
As, in addition to mobile subscriber's client-side information, mobile subscriber's network link information and mobile subscriber spy during behavior act generation
Reference ceases；
S3, mobile subscriber's historical behavior to each convolution user are recorded, and the time order and function occurred by behavior act is arranged, composition
The behavior record list of each convolution user；
Then the two-dimentional historical data unit of convolution user is built：
In the behavior record list of each convolution user, the sliding window for being N with length slides selection successively, is wrapped every time
The convolved data group of the mobile subscriber of bar containing N historical behavior record, the N bar mobile subscribers historical behavior in convolved data group is recorded
It is spliced into the two-dimentional historical data unit that two-dimensional matrix is represented；
In wherein two-dimentional historical data unit, time that behavior act occurs newest mobile subscriber's historical behavior record
Behavior act information, is used as the behavior label of the behavior label field of two-dimentional historical data unit；Simultaneously, it is necessary to two-dimentional history number
The action message recorded according to newest mobile subscriber's historical behavior of the time of the behavior act generation in unit is emptied, and is filled
Default value 0；
S4, two-dimentional historical data unit and its behavior label by all convolution users, are used as input data, input to convolution god
Through carrying out behavior prediction model training in network, the parameter of convolutional neural networks behavior prediction model is obtained, and then obtains convolution
Neutral net behavior prediction model；
S5, the mobile subscriber's historical behavior record number for obtaining targeted customer, mobile subscriber's historical behavior record number of targeted customer
If greater than or equal to N-1, then targeted customer is that convolution predicts targeted customer, and otherwise targeted customer uses for other prediction targets
Family；
The time that the current request time of S6, acquisition convolution prediction targeted customer occurs as behavior act, obtain convolution prediction
The current mobile subscriber's client-side information of targeted customer, mobile subscriber's network link information and mobile subscriber's characteristic information, generation
Mobile subscriber's historical behavior record of one current convolution prediction targeted customer to be predicted, this record, which does not include, will currently predict
Behavior label；
Targeted customer is predicted convolution, and the N-1 bar mobile subscribers historical behavior that the time of housing choice behavior action generation is nearest is recorded,
And current mobile subscriber's historical behavior record to be predicted, in chronological sequence, composition includes N bar mobile subscribers historical behavior note
The two dimension target user data cell of record；The nearest N-1 bar mobile subscriber's historical records of wherein selected action time of origin
Contain the behavior label at respective record time point；
S7, by above-mentioned two dimension target user data cell input convolutional neural networks behavior prediction model in, obtain convolution prediction
Targeted customer takes the probability of each behavior label.
2. mobile subscriber's behavior prediction method according to claim 1, it is characterised in that gone through in step S1 to mobile subscriber
History data carry out characteristic processing, including data cleansing, Missing Data Filling, quantize and data normalized；Quantize and refer to
The identifier that Hash coding obtains integer is carried out to character or character string type data.
3. mobile subscriber's behavior prediction method according to claim 1, it is characterised in that history sliding window size in step S2
N is set as a fixed value or determined by algorithm.
4. mobile subscriber's behavior prediction method according to claim 1, it is characterised in that history sliding window size N values are
10 or take all mobile subscribers mobile subscriber's historical behavior record number two quantiles.
5. mobile subscriber's behavior prediction method according to claim 1, it is characterised in that slide and refer to successively in step S3
In the behavior record list of convolution user, sliding window moves the mobile use of last in 1 step, sliding window one from the front to the back
The new mobile subscriber's historical behavior record of increase by one, wherein sliding window before family historical behavior record grand window, sliding window
Mouth moves forward 1 step, until the behavior record list of traversal convolution user.
6. mobile subscriber's behavior prediction method according to claim 1, it is characterised in that behavior prediction is carried out in step S4
Model training refers to carry out convolutional neural networks model training using back-propagation algorithm.
7. mobile subscriber's behavior prediction method according to claim 1, it is characterised in that convolutional neural networks in step S4
It is defeated that behavior prediction model includes convolutional layer, pond layer, optional regularization layer, intensive full Connection Neural Network layer, SoftMax
Go out layer；Specifically, convolutional neural networks behavior prediction model include input layer, the first convolutional layer, Batch_normalize layers,
Second convolutional layer, pooling layers of the first Max, the first Dropout layers, the 3rd convolutional layer, pooling layers of the 2nd Max, second
Dropout layers and output layer；The convolution kernel size of wherein the first convolutional layer is that 5x3, convolution kernel number are 20, the second convolutional layer and
The convolution kernel size of 3rd convolutional layer is that 3x3, convolution kernel number are 15；First Dropout layers and the 2nd Dropout layer conduct about
Beam layer, the numerical result come with 0.5 probability dropping preceding layer network delivery；Pooling layers of Max uses 2x2 sizes
Max is operated.
8. mobile subscriber's behavior prediction method according to claim 7, it is characterised in that wherein convolutional neural networks behavior
Convolutional layer number in forecast model is empirically determined or possesses the unit of direct-connected path by introducing, and semi-automatedly determines
It is fixed.
9. mobile subscriber's behavior prediction method according to claim 1, it is characterised in that convolutional neural networks behavior prediction
Model is output as multiple probable values to each two-dimentional historical data unit, represents the behavior label of the two-dimentional historical data unit
Domain obtains the probability of each behavior label.
10. mobile subscriber's behavior prediction method according to claim 1, it is characterised in that to other predictions targeted customer,
Predict that other predictions targeted customer takes the probability of each behavior label using the second forecast model；Second forecast model refers to stream
One kind in row degree model, linear prediction model, or popularity model and linear prediction model combination.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710243498.8A CN107153887A (en) | 2017-04-14 | 2017-04-14 | A kind of mobile subscriber's behavior prediction method based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710243498.8A CN107153887A (en) | 2017-04-14 | 2017-04-14 | A kind of mobile subscriber's behavior prediction method based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107153887A true CN107153887A (en) | 2017-09-12 |
Family
ID=59793999
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710243498.8A Pending CN107153887A (en) | 2017-04-14 | 2017-04-14 | A kind of mobile subscriber's behavior prediction method based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107153887A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107818251A (en) * | 2017-09-27 | 2018-03-20 | 维沃移动通信有限公司 | A kind of face identification method and mobile terminal |
| CN108280511A (en) * | 2018-01-10 | 2018-07-13 | 北京掌阔移动传媒科技有限公司 | A method of network access data is carried out based on convolutional network and is handled |
| CN108846425A (en) * | 2018-05-30 | 2018-11-20 | 大连理工大学 | Mu Ke forum subject classification method based on convolutional neural networks |
| CN109272165A (en) * | 2018-09-30 | 2019-01-25 | 江苏满运软件科技有限公司 | Register probability predictor method, device, storage medium and electronic equipment |
| CN109785034A (en) * | 2018-11-13 | 2019-05-21 | 北京码牛科技有限公司 | User's portrait generation method, device, electronic equipment and computer-readable medium |
| CN110826683A (en) * | 2018-08-07 | 2020-02-21 | 北京京东尚科信息技术有限公司 | Method and device for predicting user behavior occurrence probability |
| CN111709766A (en) * | 2020-04-14 | 2020-09-25 | 中国农业银行股份有限公司 | User behavior prediction method and device, storage medium and electronic equipment |
| CN112148352A (en) * | 2020-09-16 | 2020-12-29 | 京东数字科技控股股份有限公司 | Component configuration method, device, equipment and computer readable medium |
| WO2023029853A1 (en) * | 2021-09-02 | 2023-03-09 | 中兴通讯股份有限公司 | Model training method, data processing method, electronic device, and computer-readable storage medium |
-
2017
- 2017-04-14 CN CN201710243498.8A patent/CN107153887A/en active Pending
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107818251A (en) * | 2017-09-27 | 2018-03-20 | 维沃移动通信有限公司 | A kind of face identification method and mobile terminal |
| CN108280511A (en) * | 2018-01-10 | 2018-07-13 | 北京掌阔移动传媒科技有限公司 | A method of network access data is carried out based on convolutional network and is handled |
| CN108846425A (en) * | 2018-05-30 | 2018-11-20 | 大连理工大学 | Mu Ke forum subject classification method based on convolutional neural networks |
| CN110826683A (en) * | 2018-08-07 | 2020-02-21 | 北京京东尚科信息技术有限公司 | Method and device for predicting user behavior occurrence probability |
| CN109272165A (en) * | 2018-09-30 | 2019-01-25 | 江苏满运软件科技有限公司 | Register probability predictor method, device, storage medium and electronic equipment |
| CN109785034A (en) * | 2018-11-13 | 2019-05-21 | 北京码牛科技有限公司 | User's portrait generation method, device, electronic equipment and computer-readable medium |
| CN111709766A (en) * | 2020-04-14 | 2020-09-25 | 中国农业银行股份有限公司 | User behavior prediction method and device, storage medium and electronic equipment |
| CN111709766B (en) * | 2020-04-14 | 2023-08-18 | 中国农业银行股份有限公司 | User behavior prediction method and device, storage medium and electronic equipment |
| CN112148352A (en) * | 2020-09-16 | 2020-12-29 | 京东数字科技控股股份有限公司 | Component configuration method, device, equipment and computer readable medium |
| WO2023029853A1 (en) * | 2021-09-02 | 2023-03-09 | 中兴通讯股份有限公司 | Model training method, data processing method, electronic device, and computer-readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107153887A (en) | A kind of mobile subscriber's behavior prediction method based on convolutional neural networks | |
| CN109767301B (en) | Recommendation method and system, computer device and computer readable storage medium | |
| Kollat et al. | A new epsilon-dominance hierarchical Bayesian optimization algorithm for large multiobjective monitoring network design problems | |
| CN105119809B (en) | dynamic information display method and device | |
| US8935303B2 (en) | Method and system of optimizing a ranked list of recommended items | |
| CN110378434A (en) | Training method, recommended method, device and the electronic equipment of clicking rate prediction model | |
| CN109857871B (en) | User relationship discovery method based on social network mass contextual data | |
| CN111339433A (en) | Information recommendation method and device based on artificial intelligence and electronic equipment | |
| CN107391542A (en) | A kind of open source software community expert recommendation method based on document knowledge collection of illustrative plates | |
| CN111611488B (en) | Information recommendation method and device based on artificial intelligence and electronic equipment | |
| CN101359995A (en) | Method and apparatus providing on-line service | |
| CN104050258A (en) | Group recommendation method based on interest groups | |
| CN111125519B (en) | User behavior prediction method, device, electronic equipment and storage medium | |
| CN112634018A (en) | Overdue monitoring method for optimizing recurrent neural network based on ant colony algorithm | |
| CN106919370A (en) | The acquisition methods and device of a kind of object attribute values | |
| CN107516235A (en) | Commodity preference predictor method and device | |
| CN109543818A (en) | A kind of link evaluation method and system based on deep learning model | |
| CN110502702A (en) | User's behavior prediction method and device | |
| CN103198432B (en) | Detection method and detection system of network groups in online social network | |
| CN117315331A (en) | Dynamic graph anomaly detection method and system based on GNN and LSTM | |
| JP6470965B2 (en) | Advertisement selection device, advertisement selection method and program | |
| CN110502701B (en) | Friend recommendation method, system and storage medium introducing attention mechanism | |
| CN108122029A (en) | A kind of recommendation method and device of camera special effect | |
| CN113254795A (en) | Training method and device for recommendation model | |
| CN114036405A (en) | Social contact recommendation method and system based on graph convolution network |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170912 |
|
| RJ01 | Rejection of invention patent application after publication |