CN109977029A - A kind of training method and device of page jump model - Google Patents
A kind of training method and device of page jump model Download PDFInfo
- Publication number
- CN109977029A CN109977029A CN201910281458.1A CN201910281458A CN109977029A CN 109977029 A CN109977029 A CN 109977029A CN 201910281458 A CN201910281458 A CN 201910281458A CN 109977029 A CN109977029 A CN 109977029A
- Authority
- CN
- China
- Prior art keywords
- page
- training
- sample
- model
- jump
- 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
- 238000012549 training Methods 0.000 title claims abstract description 257
- 238000000034 method Methods 0.000 title claims abstract description 121
- 238000012360 testing method Methods 0.000 claims abstract description 33
- 230000009471 action Effects 0.000 claims description 160
- 230000009191 jumping Effects 0.000 claims description 57
- 238000010276 construction Methods 0.000 claims description 55
- 238000010998 test method Methods 0.000 claims description 15
- 238000012546 transfer Methods 0.000 claims description 8
- 238000004088 simulation Methods 0.000 claims description 4
- 239000012141 concentrate Substances 0.000 abstract description 8
- 238000013112 stability test Methods 0.000 abstract description 5
- 239000013598 vector Substances 0.000 description 32
- 230000008569 process Effects 0.000 description 22
- 230000006399 behavior Effects 0.000 description 11
- 238000004422 calculation algorithm Methods 0.000 description 9
- 238000013135 deep learning Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 238000009434 installation Methods 0.000 description 6
- 241001269238 Data Species 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000001960 triggered effect Effects 0.000 description 5
- 241000282693 Cercopithecidae Species 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 238000004590 computer program Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000005194 fractionation Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000000284 resting effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000004215 lattice model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/36—Preventing errors by testing or debugging software
- G06F11/3668—Software testing
- G06F11/3672—Test management
- G06F11/3684—Test management for test design, e.g. generating new test cases
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Quality & Reliability (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
This application discloses the training methods and device of a kind of page jump model, and by the training sample set using building, training obtains page jump model.Due to training sample concentrate include can in coverage goal APP all pages sample data, thus, according to the page jump model that training sample set training obtains, all pages can be provided in covering target APP or most of the page is corresponding jumps movement.So, when carrying out the stability test of target APP using the page jump model, it can be jumped between all pages or most of page of target APP, and all pages of target APP are tested, to improve the coverage of page jump and the Test coverage degree of the APP page.
Description
Technical field
This application involves field of computer technology more particularly to a kind of training methods and device of page jump model.
Background technique
In the development process of application program (Application, App), the stability of App is in its Quality Control Links
Various unexpected crash situations may occur in stability test for one of essential test item, App, and steady
The target of qualitative test is exactly to carry out extensive dry run to App, then excavates App to the greatest extent in extensive range
Potential risk.
About the quality of App stability test effect, wherein be on the one hand embodied on the Test coverage degree of the App page, but
Existing testing tool can only carry out random operation to several layers of pages of the superficial of APP, it is easy between these superficial pages
It is jumped repeatedly, causes the coverage of the page extremely low, the true operation that cannot be close to the users to the APP page, and then lead to APP
The page coverage of test is lower.
Summary of the invention
The main purpose of the embodiment of the present application is to provide a kind of training method of page jump model, be jumped using the page
The page jump movement that revolving die type provides, can be improved the coverage of page jump.
The embodiment of the present application provides a kind of training method of page jump model, comprising:
Construct training sample set;The training sample set includes each sample for covering all pages of the target APP
Notebook data, the sample data include the sample action executed on target APP, execute initial page before the sample action
Face mark executes target pages mark after the sample action;
According to the training sample set, training obtains page jump model, and the page jump model is for providing covering
The a series of of the wholly or largely page of the target APP jump movement.
Optionally, the building training sample set includes:
Generate target APP jumps relation data, it is described jump relation data include the target APP each page it
Between jump relationship;
Using the page jump model of initial construction, relation data is jumped according to described in and provides each sample action, and lead to
It crosses and executes each sample action to traverse all pages of the target APP;
According to the page traverse path to the target APP, the training sample set is constructed.
Optionally, the target APP of generating jumps relation data, comprising:
Page jump is realized by each effective control of the homepage of the triggering target APP, to enter the master
Under the page each first from the page;
Page jump is realized from each effective control of the page by triggering described first, to enter described first from page
Under face each second from the page, until traversal has jumped all pages of the target APP;
According to each page for jumping correspondence, generate the target APP jumps relation data.
Optionally, described according to the training sample set, training obtains page jump model and includes:
According to the training sample set, the page jump model of initial construction is trained, the final page is obtained and jumps
Revolving die type;
Alternatively, being carried out according to each sample set that the training sample is concentrated to the page jump model of initial construction
Training, obtains each page and jumps submodel, constitute final page jump model.
Optionally, the page jump model to initial construction is trained, comprising:
T batch training is carried out to the page jump model of initial construction, described batch of training is used to use K sample data
Parameter update is carried out to current page jump model, T is greater than or equal to 1;
At T times after criticizing training, continue to carry out page jump model T batch training, up to page jump mould
Until type meets training termination condition.
Optionally, the sample data further include: execute the traversal the update whether sample action occurs the page later.
Optionally, the sample data further include: the corresponding financial value of the sample action;Wherein,
It is updated if executing and the traversal of the page occurring after the sample action, the corresponding financial value of the sample action is
First financial value;
If there is no before and after the update of the traversal of the page and the execution sample action after executing the sample action
The page is different, then the corresponding financial value of the sample action is the second financial value;
If there is no before and after the update of the traversal of the page and the execution sample action after executing the sample action
The page is identical, then the corresponding financial value of the sample action is third financial value;
First financial value, second financial value, the third financial value size successively successively decrease.
It is optionally, described that parameter update is carried out to current page jump model using K sample data, comprising:
The corresponding a reference value of K sample data is generated using current baseline network model, and utilizes current page
Face jumps model and generates the corresponding predicted value of K sample data;
According to K a reference value and predicted value, parameter update is carried out to current page jump model, completes primary described batch
Training;
Wherein, the model parameter of the page jump model obtained after T described batch of training is get copied to current reference net
Network model;The a reference value is to hold after executing the sample action in corresponding sample data and entering target pages by initial page
The maximum value in each total revenue that each different series movement of row generates;The predicted value be since the initial page it
The total revenue that prediction action afterwards generates.
The embodiment of the present application also provides a kind of page jump test methods, comprising:
Using page jump model trained in advance, acted according to the page operation that the page jump model provides, it is right
Target APP carries out page jump test;
Wherein, the page jump model is any embodiment using the training method of above-mentioned page jump model
The page jump model that training obtains.
Optionally, the method also includes:
Before jumping to next page from current page using the page jump model, preset page operation is utilized
Tool, in the page operation behavior with randomness of current page simulation real user, and to the page operation work
The operating time of tool carries out timing;
After timing reaches the first preset duration, the next page is jumped to.
The embodiment of the present application also provides a kind of training devices of page jump model, comprising:
Construction unit, for constructing training sample set;The training sample set includes covering owning for the target APP
Each sample data of the page, the sample data include the sample action executed on target APP, execute the sample action
Initial page mark before executes target pages mark after the sample action;
Training unit, for according to the training sample set, training to obtain page jump model, the page jump model
It covers a series of the of the wholly or largely page of the target APP for providing and jumps movement.
Optionally, the construction unit includes:
Subelement is generated, for generating the relation data that jumps of target APP, the relation data that jumps includes the target
Relationship is jumped between each page of APP;
Subelement is traversed, for the page jump model using initial construction, relation data is jumped according to described in and provides respectively
A sample action, and by executing each sample action to traverse all pages of the target APP;
Subelement is constructed, for constructing the training sample set according to the page traverse path to the target APP.
Optionally, the generation subelement, comprising:
First trigger module realizes the page for each effective control by the homepage for triggering the target APP
It jumps, to enter under the homepage each first from the page;
Second trigger module, for realizing page jump from each effective control of the page by triggering described first,
To enter described first from each second under the page from the page, until traversal has jumped all pages of the target APP;
Submodule is generated, for according to each page for jumping correspondence, generate the target APP to jump relationship number
According to.
Optionally, the training unit is specifically used for:
According to the training sample set, the page jump model of initial construction is trained, the final page is obtained and jumps
Revolving die type;
Alternatively, being carried out according to each sample set that the training sample is concentrated to the page jump model of initial construction
Training, obtains each page and jumps submodel, constitute final page jump model.
Optionally, the training unit, comprising:
Training subelement is criticized, T batch training is carried out for the page jump model to initial construction, trains for described batch and uses
In carrying out parameter update to current page jump model using K sample data, T is greater than or equal to 1;
Subelement is recycled, for after criticizing training, continuing batch training to the progress of page jump model T times at T times,
Until page jump model meets training termination condition.
Optionally, the sample data further include: execute the traversal the update whether sample action occurs the page later.
Optionally, the sample data further include: the corresponding financial value of the sample action;Wherein,
It is updated if executing and the traversal of the page occurring after the sample action, the corresponding financial value of the sample action is
First financial value;
If there is no before and after the update of the traversal of the page and the execution sample action after executing the sample action
The page is different, then the corresponding financial value of the sample action is the second financial value;
If there is no before and after the update of the traversal of the page and the execution sample action after executing the sample action
The page is identical, then the corresponding financial value of the sample action is third financial value;
First financial value, second financial value, the third financial value size successively successively decrease.
Optionally, described batch of trained subelement, comprising:
Generation module, for generating the corresponding a reference value of K sample data using current baseline network model, and
The corresponding predicted value of K sample data is generated using current page jump model;
Update module, for carrying out parameter update to current page jump model according to K a reference value and predicted value,
Complete primary described batch of training;
Wherein, the model parameter of the page jump model obtained after T described batch of training is get copied to current reference net
Network model;The a reference value is to hold after executing the sample action in corresponding sample data and entering target pages by initial page
The maximum value in each total revenue that each different series movement of row generates;The predicted value be since the initial page it
The total revenue that prediction action afterwards generates.
The embodiment of the present application also provides a kind of page jump test devices, comprising:
Test cell, for utilizing page jump model trained in advance, the page provided according to the page jump model
Face operational motion carries out page jump test to target APP;
Wherein, the page jump model is any embodiment using the training device of above-mentioned page jump model
The page jump model that training obtains.
Optionally, described device further include:
Analogue unit, for utilizing before jumping to next page from current page using the page jump model
Preset page operation tool simulates the page operation behavior with randomness of real user in the current page, and right
The operating time of the page operation tool carries out timing;
Jump-transfer unit, for jumping to the next page after timing reaches the first preset duration.
Based on the above-mentioned technical proposal, the application has the advantages that
In the training method of page jump model provided by the embodiments of the present application, by utilizing the training sample set of building,
Training obtains page jump model.Due to training sample concentrate include can in coverage goal APP all pages sample number
According to, thus, according to the page jump model that training sample set training obtains, all pages in covering target APP can be provided
Face or most of the page is corresponding jumps movement.In this way, when the stability test for using the page jump model to carry out target APP
When, it can be jumped between all pages or most of page of target APP, and all pages of target APP are carried out
Test, to improve the coverage of page jump and the Test coverage degree of the APP page.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is the application
Some embodiments for those of ordinary skill in the art without creative efforts, can also basis
These attached drawings obtain other attached drawings.
Fig. 1 is the flow chart of the training method for the page jump model that the application embodiment of the method one provides;
Fig. 2 is a kind of schematic diagram of training sample set provided by the embodiments of the present application;
Fig. 3 is a kind of structural schematic diagram for jumping relation data provided by the embodiments of the present application;
Fig. 4 is the structural schematic diagram that another kind provided by the embodiments of the present application jumps relation data;
Fig. 5 is the flow chart of the training method for the page jump model that the application embodiment of the method two provides;
Fig. 6 is a kind of structural schematic diagram for jumping relationship subdata provided by the embodiments of the present application;
Fig. 7 is the flow chart for the page jump test method that the application embodiment of the method three provides;
Fig. 8 is the structural schematic diagram of the training device for the page jump model that the application Installation practice one provides;
Fig. 9 is the structural schematic diagram for the page jump test device that the application Installation practice two provides.
Specific embodiment
To keep the purposes, technical schemes and advantages of the embodiment of the present application clearer, below in conjunction with the embodiment of the present application
In attached drawing, the technical scheme in the embodiment of the application is clearly and completely described, it is clear that described embodiment is
Some embodiments of the present application, instead of all the embodiments.Based on the embodiment in the application, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall in the protection scope of this application.
Embodiment of the method one
Referring to Fig. 1, which is the flow chart of the training method for the page jump model that the application embodiment of the method one provides.
The training method of page jump model provided by the embodiments of the present application, comprising:
S101: building training sample set.
S102: according to training sample set, training obtains page jump model, and page jump model is for providing coverage goal
The a series of of the wholly or largely page of APP jump movement.
The above are the specific execution steps of the training method of the page jump model of the offer of the application embodiment of the method one, are
It is easy to understand and explains the training method for the page jump model that the application embodiment of the method one provides, below will successively introduce
The specific embodiment of S101 and S102.
The specific embodiment of S101 is introduced first.
In S101, training sample set includes: each sample data for covering all pages of target APP, moreover, often
A sample data can indicate the jump information in target APP between the different pages.
Sample data may include: the sample action executed on target APP, execute sample action before initial page
Target pages mark after mark, execution sample action.
In addition, sample data can be described using arbitrary forms such as text, symbol, figure or vectors, for example, working as sample
When data are described using vector, the structure of sample data can be [sF,a,sL], wherein sFIndicate execute sample action it
Preceding initial page mark;sLIt indicates to execute the target pages mark after sample action;A expression executes on target APP
Sample action.
For the ease of explanation and illustration training sample set, below by by taking the sample data being indicated using vector as an example into
Row explanation.
As an example it is supposed that target APP includes 4 jump page: first page, second page, the third page and the 4th
The page, and first page can jump to second page, and second page can jump to the third page, and first page can jump
The 4th page is gone to, then training sample set may include: [1,1-2,2], [2,2-3,3] and [1,1-4,4].Wherein, 1 is first
The page iden-tity of the page;2 be the page iden-tity of second page;3 be the page iden-tity of the third page;4 be the page of the 4th page
Mark;1-2 indicates the movement that second page is jumped to from first page;2-3 indicates to jump to the third page from second page
Movement;1-4 indicates the movement that the 4th page is jumped to from first page.
It should be noted that be indicated foregoing provide page iden-tity using number and sample action using number
The specific embodiment being indicated with symbol "-", however, in this application, not limiting the representation of page iden-tity, also not
The representation of restriction movement.For example, in the embodiment of the present application, page iden-tity can be any using number, symbol, text etc.
A kind of form is indicated or is indicated using combining form, and similarly, movement can also be using number, symbol, text etc.
Any one form is indicated or is indicated using combining form.
In addition, in order to improve the integrality of training sample set and accuracy, and then improve the page jump that training obtains
The accuracy of model, present invention also provides a kind of embodiments, in this embodiment, the sample data that training sample is concentrated
Other than it may include the content of above-mentioned offer, sample data can also include: whether page to occur after executing sample action
The traversal in face updates.
Wherein, " traversal updates " can indicate the page traversal covering for occurring that increase to target APP of sample action
Degree.For example, when executing certain sample action, if had there is (i.e. sample in the target pages A after executing the sample action
Include the page A in all pages that each sample action executed before movement is covered), then it represents that target APP
The page traversal coverage do not increase, so that it is determined that the sample action without result in the page traversal update;, whereas if holding
Target pages A after the row sample action did not occurred that (each sample action executed before the sample action is covered
Do not include the page A in all pages covered), then it represents that the page traversal coverage of target APP is increased, so that it is determined that
The traversal that the sample action results in the page updates.
In the above-described embodiment, since sample data not only includes the sample action executed on target APP, executes sample
Before this movement initial page mark, execute sample action after target pages mark, further include execute sample action it
The traversal that the page whether occurs afterwards updates, thus, the training sample set including multiple sample datas can not only record target APP
All pages between jump relationship, additionally it is possible to sequence is jumped based on sample action, accurately record is to the page of target APP
Face traverses coverage.
In addition, in order to further increase the integrality of training sample set and accuracy, so that the page that training obtains
Face, which jumps model and can provide a series of of coverage goal APP more multi-page, jumps movement, and present invention also provides a kind of implementations
Mode, in this embodiment, the sample data that training sample is concentrated is other than it may include the content of above-mentioned offer, sample
Data can also include: the corresponding financial value of sample action.
Wherein, if the traversal that the page occurs after executing the sample action updates, the corresponding financial value of the sample action
For the first financial value;If being updated after executing the sample action there is no the traversal of the page and executing sample action front and back
The page it is different, then the corresponding financial value of the sample action is the second financial value;If execute after the sample action there is no
The page of the page traversed before and after updating and executing the sample action is identical, then the corresponding financial value of the sample action is third
Financial value;Moreover, the size of the first financial value, the second financial value, third financial value is successively successively decreased.
In this implementation, " financial value " is for measuring the corresponding sample data of each sample action to training sample set
Contributed value, be specifically as follows: if the traversal that the page has occurred after executing sample action updates, then it represents that the sample is dynamic
Make the page traversal coverage that corresponding sample data can be improved training sample set to target APP, determines that the sample is dynamic at this time
The contributed value for making corresponding sample data is higher, then can set the corresponding financial value of the sample action as the first financial value;
If there is no the page before and after the traversal update of the page and execution sample action is different after executing sample action, then it represents that
The corresponding sample data of the sample action can be improved training sample and concentrate the quantity for jumping movement, determine the sample action at this time
The contributed value of corresponding sample data is medium, then can set the corresponding financial value of the sample action as the second financial value;If
Executing sample action, there is no the page before and after the traversal update of the page and execution sample action is identical later, then it represents that should
Sample action is invalid action, determines that the contributed value of the corresponding sample data of the sample action is lower at this time, then can set
The corresponding financial value of the sample action is third financial value.
First financial value, the second financial value and third financial value can be preset, for example, first can be preset
Financial value is that the 1, second financial value is 0 and third financial value is -1.
In the above-described embodiment, since each sample data that training sample is concentrated further includes the corresponding receipts of sample action
Benefit value, moreover, the financial value is for measuring the corresponding sample data of each sample action to the contributed value of training sample set, because
And the training sample set including multiple sample datas can also accurately record the financial value of each sample data, so as to subsequent
Page jump model can more accurately be trained using the sample data for including earned value.
In addition, present invention also provides a kind of embodiments, in this embodiment, sample data in addition to may include on
Other than the content for stating offer, sample data can also include: the state vector before executing sample action, and executes sample and move
State vector after work.
Wherein, " state vector before sample action is executed " for indicating during traversing target APP, execute sample
Page ergodic state information possessed by before this movement.
" executing the state vector after sample action " it is dynamic to execute sample for indicating during traversing target APP
Page ergodic state information possessed by after making.
" state vector " can be determined according to the page total quantity of target APP and the ergodic process of target APP, moreover, " shape
State vector " can be indicated using various ways.
For example, " state vector " can be indicated using N-dimensional vector, wherein N is the page total quantity of target APP;And
And the 1st dimension to the N-dimensional of " state vector " respectively corresponds the 1st page to the n-th page, when the value of the i-th dimension of state vector
When being 1, then it represents that currently resting on i-th of page of target App;When the value of the i-th dimension of state vector is 0, then
Indicate that i-th of page of target App had been traversed;When the value of the i-th dimension of state vector is -1, then it represents that target App's
I-th of page was not traversed.
It, below will be with the training sample set shown in Fig. 2 including state vector for the ease of explanation and understanding state vector
For be illustrated.
Assuming that target APP includes 6 pages: homepage, first page to the 5th page, and homepage is identified as " 0 ",
The mark of first page to the 5th page is respectively " 1 " to " 5 ".
Training sample set shown in Fig. 2 includes 5 sample datas, and each row of data indicates a sample data, moreover, each
Initial page before sample data includes: the sample action executed on target APP, executes sample action identifies, executes sample
The state vector before target pages mark, the corresponding financial value of sample action, execution sample action after movement, and hold
State vector after row sample action.
For the ease of explanation and understanding sample data shown in Fig. 2, below by by taking the sample data 201 in Fig. 2 as an example into
Row is specific to be introduced.
As an example, as shown in Fig. 2, sample data 201 include: " [1-1-1-1-1-1] ", " 0-3 ", " 1 " and
Four parts of " [0-1-1 1-1-1] ";Moreover,
First part: " [1-1-1-1-1-1] " indicates to rest on homepage before executing sample action;
Second part: " 0-3 " indicates that the sample action executed is to jump to the third page from homepage, wherein " 0 " is to hold
Initial page mark before row sample action, and " 3 " are the target pages mark after executing sample action;
Part III: " 1 " indicates that the financial value of the sample action executed is 1;
Part IV: resting on the third page after " [0-1-1 1-1-1] " expression execution sample action, and
Homepage had been traversed.
In the above-described embodiment, the state vector before further including execution sample action due to each sample data, with
And the state vector after execution sample action, thus, the training sample set including multiple sample datas can be accurately complete
Reflect the ergodic process to target APP, the subsequent ergodic process that can be provided according to training sample concentration is provided, the page is jumped
Revolving die type optimizes, to be conducive to improve the page traversing capabilities of page jump model.
Foregoing provide the related content of training sample set, training sample set provided by the present application includes multiple sample numbers
According to moreover, multiple sample data can indicate the jump information in target APP between each page, so that training sample
This collection is able to record the jump information between all pages of target APP, and then improves training sample set to target APP's
The coverage rate of all pages.
Next, this application provides a kind of specific embodiments of building training sample set, in this embodiment,
S101 can specifically include S1011-S1013:
S1011: generating target APP and jump relation data, this jump relation data include target APP each page it
Between jump relationship.
It jumps relation data and jumps relationship between the different pages for recording in target APP, moreover, jumping relationship number
According to different expression ways can be used to be described.For example, jumping relation data can be described using textual form,
It can be described using tree figure.
Relation data is jumped with explanation in order to facilitate understanding, successively will jump relationship below with what is described using textual form
It data and is introduced for relation data using jumping for tree figure description.
The related content for jumping relation data described using textual form is introduced first.
As an example, when using textual form to jumping relation data and being described, target APP is corresponding to jump pass
Coefficient using text data as shown in Figure 3 according to that can be indicated, moreover, each row of data in Fig. 3 can indicate two
Relationship is jumped between different interfaces.
For example, text data 301 is used to indicate the phase for jumping relationship between homepage and first page in Fig. 3
Close information, wherein " 0 " is the page iden-tity of homepage;" 1 " is the page iden-tity of first page;" 0-1 " can indicate homepage
Jump to the movement of first page;" [652,73] " and " [700,121] " can indicate corresponding first control of triggering " 0-1 " movement
The location information of part, concrete meaning be, the first control, which is located at, to be surrounded by point [652,73] and point [700,121] on homepage
In rectangular area.
It should be noted that above-mentioned is to jumping what relation data was described by taking textual form shown in Fig. 3 as an example, so
And target APP can be not only described using textual form shown in Fig. 3 to jumping relation data in this application, may be used also
To be described using other textual forms to relation data is jumped, the embodiment of the present application is not specifically limited in this embodiment.
The above are the related contents for jumping relation data described using textual form.
The related content for jumping relation data described using tree figure is described below.
As another example, when using tree figure to jumping relation data and being described, the corresponding jump of target APP
Transferring the registration of Party membership, etc. from one unit to another data can be indicated using tree figure as shown in Figure 4;Moreover, each circle in Fig. 4 is for indicating
Each page of target APP, " → " in Fig. 4 jump relationship between the different pages in target APP for describing.
For example, " 0 " is the page iden-tity of homepage in Fig. 4;" 1 " is respectively the page iden-tity of first page to " 15 "
To the page iden-tity of the 15th page;First arrow 401 is for being expressed as the movement that homepage jumps to first page.
It should be noted that above-mentioned is by taking tree figure shown in Fig. 4 as an example to jumping what relation data was described,
However, target APP can be not only described using tree figure shown in Fig. 4 to relation data is jumped in this application,
It can also be described using other tree figures to relation data is jumped, the embodiment of the present application does not do specific limit to this
It is fixed.
The above are the related contents for jumping relation data described using tree figure.
Based on the above-mentioned related content for jumping relation data, the embodiment of the present application provides generation and jumps the one of relation data
Kind embodiment, in this embodiment, S1011 can specifically include S1011a-S1011c:
S1011a: realizing page jump by each effective control of the homepage of triggering target APP, to enter homepage
Under face each first from the page.
Effective control refers to the control that page jump is able to carry out when triggering, that is to say, that if the first control is triggered,
Current page is enabled to jump to other pages other than current page, then the first control is effective control;If the
One control is triggered, so that current page can not jump to other pages other than current page, then the first control is not
Effective control.
Include multiple effective controls on the homepage of target APP, moreover, each effectively control corresponding one jump it is dynamic
Make, so that the movement that other pages are once jumped to by homepage can be carried out when each effective control is triggered, thus
Realize jumping between homepage and other pages.
First is to refer to act the page jumped to from homepage by one from the page, for example, when homepage can
When jumping to first page by " 0-1 " movement, then it represents that first page is one first under homepage from the page.
As an example it is supposed that on the homepage of target APP include 5 controls: the first control, the second control, third control,
4th control and the 5th control, moreover, the corresponding movement of the first control is " 0-1 ", the corresponding movement of the second control is " 0-2 ",
The corresponding movement of third control is " 0-3 ", and the corresponding movement of the 4th control is " 0-4 ", and the corresponding movement of the 5th control is " 0-
5 ", then S1011a is specifically as follows: when triggering the first control in target APP on homepage, then can enter first page;
When triggering the second control in target APP on homepage, then it can enter second page;When homepage in triggering target APP
On third control when, then can enter the third page;It, then can be with when triggering four control in target APP on homepage
Into the 4th page;When triggering five control in target APP on homepage, then it can enter the 5th page.
S1011b: realizing page jump from each effective control of the page by triggering first, to enter first from page
Under face each second from the page, until traversal has jumped all pages of target APP.
Due to including multiple effective controls on each page of target APP, moreover, each effectively control is one corresponding
Movement is jumped, so that can carry out once jumping to other pages by current page when each effective control is triggered
Movement, to realize jumping between current page and other pages, moreover, it is similar to jump mode between the different pages,
Thus, by first provided from page jump to second from the embodiment of the page and S1011a by homepage jump to first from
The embodiment of the page is identical, and for the sake of brevity, details are not described herein.
S1011c: according to each page for jumping correspondence, generate target APP jumps relation data.
As an implementation, S1011c is specifically as follows: when it is each jump generation when, then record it is each jump front and back
The corresponding page;And when traversal has jumped all pages of target APP, then according to each correspondence that jumps of record
The page, generate target APP jumps relation data.
As an example, S1011c is specifically as follows when target APP includes 15 pages: when target pages jump to the
When one page, the movement of " 0-1 " is recorded;When target pages jump to second page, the movement of " 0-2 " is recorded;Work as page object
When face jumps to the third page, the movement of " 0-3 " is recorded;When target pages jump to four pages, record " 0-4 " is moved
Make;When target pages jump to five pages, the movement of " 0-5 " is recorded;When first page jumps to six pages, record
The movement of " 1-6 ";When first page jumps to seven pages, the movement of " 1-7 " is recorded;When first page jumps to page eight
When face, the movement of " 1-8 " is recorded;When the 6th page jump to nine pages, the movement of " 6-9 " is recorded;When the 8th page is jumped
When going to ten pages, the movement of " 8-10 " is recorded;When the 8th page jump to 11 page, record " 8-11 " is moved
Make;When second page jumps to 12 page, the movement of " 2-12 " is recorded;When third page jump to the 13rd page
When, record the movement of " 3-13 ";When the 4th page jump to 14 page, the movement of " 4-14 " is recorded;When the 5th page
When jumping to 15 page, the movement of " 5-15 " is recorded;Then, according to the movement recorded, target shown in Fig. 4 is generated
APP's jumps relation data.
It should be noted that the application can by it is each jump movement occur before and after carry out respectively screenshotss, protect
It deposits and analyzes, jump the corresponding control of movement in order to obtain each page for jumping correspondence and this.
The above are the specific embodiments of S1011, in this embodiment, firstly, since the homepage of target APP into
The movement of row page jump, and page jump acts specifically: by each effective control on triggering current page, to realize
Other pages other than current page are jumped to from current page;Then, it repeatedly triggers effective on the current page
Control carries out page jump, until all pages that traversal has jumped target APP then stop carrying out page jump;Finally, according to
Each page for jumping correspondence, generate target APP jumps relation data.In this way, since multiple page jump acts energy
All pages of target APP are enough traversed, thus, the page for acting correspondence according to each page jump is generated to jump pass
Coefficient evidence can completely record the jump information between all pages of target APP, to improve the jump of target APP
Transfer the registration of Party membership, etc. from one unit to another the integrality and accuracy of data.
S1012: using the page jump model of initial construction, providing each sample action according to relation data is jumped, and
By executing each sample action to traverse all pages of target APP.
Page jump model, using relation data is jumped, can obtain each sample action according to the page info of input
Corresponding executable probability;Wherein, probability can be performed for describing a possibility that each sample action is performed size, if can hold
The row probability the big, indicates that a possibility that sample action is performed is bigger, if executable probability is smaller, then it represents that the sample is dynamic
A possibility that being performed is smaller.
For example, it is assumed that the relation data that jumps of target APP is tree figure shown in Fig. 4, and to page jump mould
When type inputs the information of homepage, then the executable probability of first page to the 15th page can be obtained, so as to according to
First page determines sample action to be executed on homepage to the executable probability of the 15th page.
It should be noted that executable probability can be determined according to the corresponding financial value of sample action, such as from homepage
The financial value of first page to the 15th page is jumped to respectively, and the method for determination of financial value refers to above-mentioned related introduction;When
So, it can also be determined according to the factor that other influences jump speed, the embodiment of the present application is not especially limited this.
In addition, the page jump model of initial construction can be the page jump model based on original model parameter building,
Wherein, original model parameter can be preset, for example, original model parameter can be set according to application scenarios;Moreover, initial
The page jump model of building is also possible to the page jump model constructed based on other modes.
The relevant knowledge of page jump model based on above-mentioned introduction, the embodiment of the present application also provides one kind of S1012
Specific embodiment, in this embodiment, S1012 are specifically as follows:
S1012a: the relevant information of current page is input in the page jump model of initial construction, obtains each sample
The corresponding executable probability of this movement.
S1012b: the corresponding executable probability of more each sample action obtains maximum executable probability, and executes maximum
The corresponding sample action of probability can be performed and carry out page jump.
S1012c: judging whether to have traversed all pages of target APP, if so, executing S1012e;If it is not, then executing
S1012d。
S1012d: it using the page after the maximum executable corresponding sample action of probability of execution as current page, returns
Execute S1012a.
S1012e: stop traversal target APP.
In the specific embodiment of the S1012 of above-mentioned offer, it can realize by the movement of multiple page jump to target
The traversal of all pages in APP, moreover, the process of each page jump specifically: according to the page jump model of initial construction
The corresponding executable probability of each sample action of output, determination executes movement on current page, and executes the determination
Execution movement carries out page jump.In this way, since each page jump selects the sample action of executable maximum probability to carry out
It jumps, thus, it is possible to which all pages for improving traversal target APP jump speed, to improve the building effect of training sample set
Rate.
S1013: according to the page traverse path to target APP, training sample set is constructed.
As an example it is supposed that target APP includes: homepage, first page to the 5th page, moreover, to the page of target APP
Face traverse path the are as follows: when homepage → third page → the 4th page → five pages → first page → second page, then
S1013 specifically: according to the page traverse path to target APP, construct training sample set (for example, training sample set can wrap
Include 5 sample datas shown in Fig. 2).
Since the page traverse path to target APP can be recorded accurately to the visit between the pages different in target APP
Sequence is asked, so that not only including jumping pass between the different pages according to the training sample set that the page traverse path constructs
It is the access order further comprised between the pages different in target APP, thus, in the process tested target APP
In, the page jump model obtained according to the training sample set can access according to the access order that training sample is concentrated,
It carries out jumping movement from the same page to another page so as to avoid being repeated as many times, and then improves page jump mould
Traversal efficiency of the type to all pages of target APP.In addition, also due to the training sample set that is constructed according to the page traverse path
It include the access mode to the pages all in target APP, thus, during testing target APP, according to the instruction
The page jump model that white silk sample set obtains can traverse all pages in target APP, to improve page jump model
Coverage is traversed to the page of target APP.
The above are the specific embodiments of S101 provided by the embodiments of the present application, in this embodiment, can be first with
The page jump model of initial construction provides each sample action according to the relation data that jumps generated, is executed respectively with will pass through
A sample action is to traverse all pages of target APP;Further according to the page traverse path to target APP, training sample is constructed
Collection.In this way, ensure that the integrality and accuracy of training sample set, own to improve training sample set to target APP
The representativeness of the page, and then be conducive to improve the subsequent page jump model obtained according to the training sample set to target APP's
The page traverses coverage.
The specific embodiment of S102 is described below.
In one of S102 embodiment, S102 is specifically as follows: according to training sample set, to the page of initial construction
Face jumps model and is trained, and obtains final page jump model.
In the above-described embodiment, the page jump model of initial construction can be instructed using a variety of preset algorithms
Practice.Wherein, preset algorithm can be preset, for example, preset algorithm can be deeply study (Deep
Reinforcement Learning, DRL) algorithm.
DRL algorithm can be by first carrying out stochastical sampling to training sample set, and utilizes the sample data of sampling to the page
It jumps model and is iterated training, so that page jump model is optimal state.Wherein, optimum state refers to page jump mould
Type can traverse all pages of target APP with least step number.
The decision-making capability of the sensing capability of deep learning and intensified learning can be combined by DRL algorithm, thus, DRL
Algorithm relates to intensified learning process and deep learning process, for the ease of explanation and understanding DRL algorithm, will successively introduce below
Intensified learning process and deep learning process.
The related content of intensified learning process is introduced first.
During intensified learning, it is described for the ease of the jump procedure between the different pages to target APP, it can
With will traverse target APP all pages procedural abstraction be a Markovian decision process;Moreover, Markov decisior process
Journey may include: state space S, motion space A, delay reward parameter value γ and jump the instant reward R after movement executes.
Wherein, the different moments that " state space S " is used to describe during traversing all pages of target APP are had
Some page ergodic state information, and status information can be using state vector form as shown in Figure 2 (for example, [1-1-
1-1-1-1] it is a state in S) it is indicated.
" motion space A " include: traverse target APP all pages during be related to it is all jump movement,
Moreover, jumping movement can be described in different ways.It can be using " 0-3 " as shown in Figure 2 for example, jumping movement
Representation be described.
" postponing reward parameter value γ " is for measuring the subsequent influence for jumping movement to movement generation is currently jumped, Er Qieyan
Reward parameter value γ can be preset late, for example, delay reward parameter value γ can be redefined for 0.9.
" jumping the instant reward R after movement executes " is for describing during traversing all pages of target APP, often
It is a to jump the reward of acquisition after movement executes, moreover, jumping the instant reward R after movement executes can preset.As
Example, can be in advance using the above-mentioned financial value provided as the instant reward R after jumping movement execution.
In addition, during intensified learning, it is also necessary to according to each relevant information for jumping movement, find one optimal time
Path is gone through, so that traversing the efficiency highest of all pages in target APP.
For the ease of the search procedure of the optimal traverse path of explanation and understanding, below by with include N number of page target APP
For be illustrated.
As an example it is supposed that target APP includes that N number of page and M jump movement, then " state space " S includes the 1st shape
State s0To n-th state sN, " motion space A " includes the 1st movement a0A is acted to m-thM, then the target in the search procedure
Function is as follows:
maxQ(s0)=maxE [R (s0,a0)+γ·R(s1,a1)+γ2·R(s2,a2)+...] (1)
In formula, maxQ (s0) indicate state s0Maximum return, moreover, state s0Maximum return correspond to state
s0Optimal traverse path of the corresponding page as traversal starting point, the optimal traverse path specifically: from state s0The corresponding page
(for example, homepage) starts, and successively jumps to state s1The corresponding page, s2The corresponding page, s3The corresponding page ... ...,
sN-1The corresponding page;MaxE [] indicates to obtain the function of optimal traverse path;R(s0,a0) indicate in state s0The corresponding page
Upper execution acts a0When the instant reward that obtains;R(si, ai) indicate in state siExecution acts a on the corresponding pageiWhen obtain
Immediately reward;γ is delay reward parameter value;a0Expression state s0Corresponding page jump is to state s1It is corresponding when the corresponding page
Jump movement;aiExpression state siCorresponding page jump is to state si+1It is corresponding when the corresponding page to jump movement;N is indicated
The total quantity of the page in target APP;M indicates the total quantity that movement is jumped in target APP.
It should be noted that γ can guarantee maxQ (s0) value be Finite Number, in order to it is subsequent can be according to maxQ (s0)
Value determine optimal traverse path.
Furthermore it is also possible to which formula (1) is carried out linear transformation, it is as follows to obtain graceful (bellman) equation of Bell:
Q(s0, a0)=R (s0,a0)+γ·maxQ(s1) (2)
Wherein, Q (s0,a0) indicate in state s0Execution acts a on the corresponding page0When the total revenue that obtains;R(s0,a0)
It indicates in state s0Execution acts a on the corresponding page0When the instant reward that obtains;γ is delay reward parameter value;maxQ(s1)
It indicates to work as state s0Corresponding page jump is to state s1When the corresponding page, with state s1The corresponding page as starting point not
Carry out maximum return.
Based on above content it is found that during the determination of optimal traverse path, need to find s0It is optimal under state to move
Make a0, but due to current optimal movement a0The not necessarily movement a of global optimum0, thus a0Selection again cannot be only
Only consider current optimal, it is also necessary to consider the subsequent influence for jumping and acting to movement is currently jumped.In this way, in this application,
It needs to reuse formula (2) to be iterated each state in ergodic process, it is corresponding to repeat to update each movement
Aggregate earnings value can then determine the optimal movement under each state, thus root when the convergence of each movement corresponding aggregate earnings value
According to the optimal movement, obtain with state s0Optimal traverse path of the corresponding page as traversal starting point.
It should be noted that each movement corresponding aggregate earnings value convergence can specifically refer to: neural network parameter is constant,
Or, the output content of neural network is constant when inputting identical content.
The above are the related contents of intensified learning process, during intensified learning, need the traversal according to target APP,
The corresponding aggregate earnings value of ergodic process of target APP is generated, can determine target according to the aggregate earnings value so as to subsequent
The optimal traverse path of APP.
The related content of deep learning process is described below.
Size due to " state space S " will exponentially increase with the page sum of target APP, so that " state is empty
Between the corresponding aggregate earnings value table of S " also exponentially increase, for example, when target APP include N number of jump page and M jump it is dynamic
Make, then " state space S " corresponding aggregate earnings value table includes 2NM total revenue, thus, each possibility can not be gone out by exhaustion
The mode of each movement under the state of appearance and the state calculates the corresponding total revenue of each state being likely to occur
Value.
In order to solve this problem, the application will use deep neural network, concentrate to the training sample that S101 is obtained
Sample data is fitted, and obtains page jump model, enables page jump model to the movement total revenue of input state
Value is predicted, is obtained the execution under input state and is each jumped movement total revenue obtained.Thus, when target APP includes N
A page and M when jumping movement, which can be predicted according to the N-dimensional state vector of input, obtain the N
It ties up the corresponding M of state vector and ties up total revenue vector, wherein every one-dimensional representation in total revenue vector executes often under current state
It is a to jump movement total revenue obtained.In this way, by page jump model can to it is stateful under movement aggregate earnings value
It is predicted, thus, the application is not necessarily to each movement under the state and the state that are each likely to occur out by exhaustion
Mode calculates the corresponding aggregate earnings value of state that is each likely to occur, it is only necessary to page jump mode input N-dimensional state vector,
The corresponding M dimension total revenue vector of the N-dimensional state vector can be obtained, so that the computational efficiency of movement aggregate earnings value is improved,
And memory space is saved.
In addition, will be trained in conjunction with baseline network model to page jump model during deep learning;Wherein,
Baseline network model is identical with the structure of page jump model, for example, being the deep neural network that several layers connect entirely;And
And baseline network model and the corresponding hidden layer of page jump model and each hidden layer number of nodes can concurrency adaptation, example
Such as, the network parameter of page jump model can be copied to base after updating several times by the network parameter of page jump model
Pseudo-crystalline lattice model.It should be noted that how baseline network model to be combined to be trained page jump model, it will be subsequent interior
It describes in detail in appearance.
The above are the related contents of deep learning process, during deep learning, can use deep neural network pair
Page jump model is trained, and page jump model is predicted according to the N-dimensional state vector of input, is somebody's turn to do
The corresponding M of N-dimensional state vector ties up total revenue vector, to improve the computational efficiency of movement aggregate earnings value, and has saved storage
Space.
The related content of DRL algorithm based on above-mentioned introduction, the embodiment of the present application also provides the instructions of page jump model
Practice a kind of embodiment of method, in this embodiment, S102 is specifically as follows: first to the page jump model of initial construction
Carry out T batch training, this batch training for using K sample data (sample data of training sample concentration in S101) to working as
Preceding page jump model carries out parameter update;After criticizing training, continue to carry out page jump model T times when at T times
Batch training, until until page jump model meets trained termination condition.
Wherein, T can be preset, and T >=1.For example, T can be redefined for 100.
For the ease of the S102 embodiment of the above-mentioned offer of explanation and understanding, will be illustrated so that T is greater than 1 as an example below.
As an implementation, when T is greater than 1, then S102 can specifically include S1021-S1022:
S1021: using the 1st batch of K sample data, carrying out parameter update to the page jump model of initial construction, complete
At the 1st time batch of training, the page jump model after the 1st time batch of training is obtained.
K can be preset, for example, K can be set previously according to application scenarios;Moreover, because K sample data be from
Training sample concentrates acquisition, thus, K≤training sample concentrates the total number of sample data, and K is positive integer.
K sample data can select according to preset rules from training sample concentration;Wherein, preset rules can be pre-
First set, for example, concentrate the Data Identification of each sample data successively to choose K sample data according to training sample, alternatively, from
Training sample concentrates K sample data of random selection.
As an implementation, S1021 can specifically include S1021a-S1021c:
S1021a: the corresponding a reference value of K sample data is generated using current baseline network model.
The a reference value is to hold after executing the sample action in corresponding sample data and entering target pages by initial page
The maximum value in each total revenue that each different series movement of row generates, wherein initial page here is corresponding sample number
The page of initial page mark meaning in, the goal page are that the target pages mark in corresponding sample data is signified
The page.In addition, what the model parameter of the baseline network model in the 1st time batch of training can be obtained with parameter initialization.
S1021b: the corresponding predicted value of K sample data is generated using current page jump model.
The predicted value be since initial page after prediction action generate total revenue, wherein here just
The initial page referred in the beginning page and S1021a is identical.
S1021c: according to K a reference value and predicted value, parameter update is carried out to current page jump model, completes the 1st
Secondary batch of training, and obtain the page jump model that the 1st time batch of training generates.
As an implementation, S1021c is specifically as follows: according to K a reference value and predicted value, using formula (3),
Obtain the corresponding penalty values loss of current page jump model;According to penalty values loss, to current page jump model
Parameter update is carried out, completes the 1st time batch of training, and obtain the page jump model generated through the 1st time batch of training.
Wherein, loss indicates the current corresponding penalty values of page jump model;The number of K expression sample data;Qq_net
(s(l)) indicate that the sample action in first of sample data of execution corresponds to obtained predicted value;Q_net indicates that the current page is jumped
Revolving die type;R_net indicates current baseline network model;γ indicates delay reward parameter value;(R(s(l),a(l))+γ·max
QT_net(s_(l), a') and indicate that the sample action in first of sample data of execution corresponds to obtained a reference value;R(s(l),a(l)) indicate
In state s(l)Execution acts a on the corresponding page(l)When the instant reward that obtains;max QT_net(s_(l), a') and indicate state s
_(l)Maximum return (calculate detail as per formula (1));a(l)Expression state s(l)Corresponding page jump is to state s_(l)It is corresponding
The page when corresponding jump movement;A' indicates state s_(l)When corresponding page jump to next page it is corresponding jump it is dynamic
Make.
If the corresponding penalty values loss of current page jump model is smaller, then it represents that current page jump model
The ability for traversing the page in target APP is higher, whereas if the corresponding penalty values loss of current page jump model is bigger,
Then indicate that the ability of the page in the traversal target APP of current page jump model is weaker, so, it can be based on the big of loss value
It is small, parameter update is carried out to current page jump model, to promote the traversing capabilities of page jump model, at this point, completing the 1st
Secondary batch of training.
It should be noted that do not have between S1021a and S1021b it is fixed execute sequence, S1021a can be executed sequentially
And S1021b, S1021b and S1021a also can be executed sequentially, may also be performed simultaneously S1021a and S1021b.
S1022: using the 2nd batch of K sample data, the page jump model obtained after the 1st time batch of training is carried out
Parameter updates, and completes the 2nd time batch of training, obtains the page jump model after the 2nd time batch of training.
It should be noted that the specific embodiment of S1022 is identical as the specific embodiment of S1021, in order to briefly rise
See, details are not described herein.
In the manner described above, K sample data for reusing the 3rd batch, to the page jump obtained after the 2nd time batch of training
Model carries out parameter update, completes the 3rd time batch of training, obtains the page jump model ... ... after the 3rd time batch of training, in this way,
Batch training constantly is carried out to page jump model, until completing T times batch of training.
It should be noted that in the above-described embodiment, K sample data used in batch training process can be every time
It is identical, it is also possible to different, the application is not especially limited this.
Next, using the page jump model obtained after T times batch of training as the page jump model of initial construction, and
The model parameter of the page jump model obtained after T times batch of training is copied to current baseline network model, returns and executes
S1021。
In this embodiment, T times batch is carried out after training due to every, the page jump that need to will be obtained after T times batch of training
The model parameter of model is copied to current baseline network model, and baseline network model is saved through this T times batch of training
The preferably page jump model of performance afterwards, so that baseline network model reaches local optimum, subsequent to page jump
When model is trained, be conducive to the training effectiveness for further increasing page jump model, and promote the page of page jump model
Face traversing capabilities.
In this way, every complete T times batch of training, the wheel training to page jump model is just realized, more trainings in rotation can be passed through
Practice, until page jump model meets training termination condition, for example, being more than default to the exercise wheel number of page jump model
Until number or loss value are less than preset threshold, at this point, obtaining final page jump model.
The above are the specific embodiments of S102, in this embodiment, can be according to training sample set, to initial construction
Page jump model be trained, obtain final page jump model.
The above are the application embodiment of the method one provide page jump model training method specific embodiment,
In the embodiment, by the training sample set using building, training obtains page jump model.It is concentrated and is wrapped due to training sample
Included can in coverage goal APP all pages sample data, thus, jumped according to the page that training sample set training obtains
Revolving die type can provide in covering target APP all pages or most of the page is corresponding jumps movement.In this way, when using
When the page jump model carries out the stability test of target APP, can all pages of target APP or most of page it
Between jumped, and all pages of target APP are tested, thus improve page jump coverage and APP pages
The Test coverage degree in face.
It, can be according to training sample data in the training method for the page jump model that above method embodiment one provides
The sample data provided is concentrated, training obtains page jump model, enables the institute of page jump model coverage goal APP
Have and jumps movement between the page or most of page.
In addition, in order to further increase the training effectiveness of page jump model, especially when the page total quantity of target APP
When larger, training sample set can be split into multiple sample sets, in order to be trained respectively according to each sample set,
Thus, present invention also provides the training method of another page jump model, below in conjunction with attached drawing carry out specific explanations and
Explanation.
Embodiment of the method two
Embodiment of the method is second is that the improvement carried out on the basis of embodiment of the method one, for the sake of brevity, method are implemented
With identical content in embodiment of the method one in example two, details are not described herein.
Referring to Fig. 5, which is the flow chart of the training method for the page jump model that the application embodiment of the method two provides.
The training method of page jump model provided by the embodiments of the present application, comprising:
S501: building training sample set.
S502: each sample set concentrated according to training sample is trained the page jump model of initial construction,
It obtains each page and jumps submodel, constitute final page jump model.
The above are the specific execution steps of the training method of the page jump model of the offer of the application embodiment of the method two, are
It is easy to understand and explains the training method for the page jump model that the application embodiment of the method two provides, below will successively introduce
The specific embodiment of S501 and S502.
The specific embodiment of S501 is introduced first.
In S501, training sample set first can be generated using the specific embodiment that S101 is provided, be drawn according still further to default
Training sample set is then divided into multiple sample sets by divider, to enable training sample set to include multiple sample sets.
Wherein, default division rule can be preset.
Specifically, S501 may include S5011-S5015:
S5011: generate target APP jumps relation data, and jumping relation data includes between each page of target APP
Jump relationship.
The specific embodiment of S5011 and the specific embodiment of S1011 are identical, and details are not described herein.
S5012: according to default fractionation rule, will jump relation data split into it is multiple jump relationship subdata, this is jumped
Relationship subnumber includes jumping relationship between the partial page of target APP.
The default rule that splits can be preset, for example, the default rule that splits can be set previously according to application scenarios.
As an example, as shown in Figure 4 and Figure 6, S5012 is specifically as follows:, can be by Fig. 4 institute according to default fractionation rule
The relation data that jumps shown splits into six shown in fig. 6 and jumps relationship subdata: first jumps relationship subdata 601 to the 6th
Jump relationship subdata 606.
It should be noted that above-mentioned be illustrated for jumping expression way of the relation data using tree figure
, however, the method for splitting for jumping relation data of above-mentioned offer is applicable not only to use the expression way of tree figure
Relation data is jumped, applies also for jumping relation data under other any expression ways.
S5013: using the page jump model of initial construction, respectively according to each jumping relationship subdata provides each sample
This movement, and by executing each sample action to traverse the partial page for jumping the corresponding target APP of relationship subdata.
As an example, when jump relationship subdata it is as shown in Figure 6 when, then S5013 is specifically as follows:
S5013a: using the page jump model of initial construction, relationship subdata 601 is jumped according to first and provides each sample
This movement, and homepage in the corresponding target APP of relationship subdata 601 is jumped by executing each sample action to traverse first
Face, first page to the 5th page.
S5013b: using the page jump model of initial construction, relationship subdata 602 is jumped according to second and provides each sample
This movement, and by executing each sample action with traverse second jump in the corresponding target APP of relationship subdata 602 first
The page, the 6th page to the 11st page.
S5013c: using the page jump model of initial construction, relationship subdata 603 is jumped according to third and provides each sample
This movement, and by executing each sample action with traverse third jump in the corresponding target APP of relationship subdata 603 second
The page and the 12nd page.
S5013d: using the page jump model of initial construction, transfer the registration of Party membership, etc. from one unit to another subdata 604 according to the forth jump and provide each sample
This movement, and transfer the registration of Party membership, etc. from one unit to another third in the corresponding target APP of subdata 604 by executing each sample action to traverse the forth jump
The page and the 13rd page.
S5013e: using the page jump model of initial construction, transfer the registration of Party membership, etc. from one unit to another subdata 605 according to the fifth jump and provide each sample
This movement, and by executing each sample action with traverse the fifth jump transfer the registration of Party membership, etc. from one unit to another in the corresponding target APP of subdata 605 the 4th
The page and the 14th page.
S5013f: using the page jump model of initial construction, relationship subdata 606 is jumped according to the 6th and provides each sample
This movement, and by executing each sample action with traverse the 6th jump in the corresponding target APP of relationship subdata 606 the 5th
The page and the 15th page.
It should be noted that S5013a to do not have between S5013f it is fixed execute sequence, can be preset according to first suitable
Sequence successively executes, and also may be performed simultaneously all steps, and the application is not especially limited this.
S5014: the corresponding page traverse path of relationship subdata is jumped according to each, constructs each sample set.
As an example, when jump relationship subdata it is as shown in Figure 6 when, then S5014 is specifically as follows: jumping pass according to first
It is that subdata 601 to the 6th jumps the corresponding page traverse path of relationship subdata 606, constructs first sample subset respectively to the
Six sample sets.
It should be noted that building no permanent order of the first sample subset to the 6th sample set, it can be according to
Two preset orders successively carry out the building of first sample subset to the 6th sample set, can also carry out first sample subset simultaneously
To the building of the 6th sample set, the application is not especially limited this.
S5015: according to multiple sample sets, training sample set is constructed.
The above are the specific embodiments of S501, in this embodiment, the embodiment that can will be provided using S101
The training sample set of building is divided into multiple sample sets, and training sample set is enabled to include multiple sample sets;It can also be with
Page jump relation data is split as multiple page jump relationship subdatas, can be closed according to each page jump so as to subsequent
It is that subdata generates each sample set respectively, and constructs training sample set according to multiple sample sets.
The specific embodiment of S502 is described below.
For the ease of explanation and understanding S502, below will in one embodiment for be illustrated.
As an implementation, when training sample set includes six sample sets: first sample subset to the 6th sample
When subset, then S502 is specifically as follows:
S5021: according to first sample subset, the page jump model of initial construction is trained, first page is obtained
Jump submodel.
S5022: according to the second sample set, the page jump model of initial construction is trained, second page is obtained
Jump submodel.
S5023: according to third sample set, the page jump model of initial construction is trained, the third page is obtained
Jump submodel.
S5024: according to the 4th sample set, the page jump model of initial construction is trained, the 4th page is obtained
Jump submodel.
S5025: according to the 5th sample set, the page jump model of initial construction is trained, the 5th page is obtained
Jump submodel.
S5026: according to the 6th sample set, the page jump model of initial construction is trained, the 6th page is obtained
Jump submodel.
S5027: submodel is jumped to the 6th page jump submodel according to first page, constitutes final page jump mould
Type.
It should be noted that S5021 to do not have between S5026 it is fixed execute sequence, can be according to third preset order
S5021 to S5026 is successively executed, also may be performed simultaneously S5021 to S5026, the application is not especially limited this.
The above are the specific embodiments of S502, in this embodiment, each sample concentrated according to training sample
Collection, is trained the page jump model of initial construction, obtains each page and jump submodel, constitute final page jump
Model.It should be noted that the training method of each page jump submodel can adopt in the specific embodiment of S502
Any training method provided in the S102 of embodiment of the method one, for the sake of brevity, details are not described herein.
The above are the application embodiment of the method two provide page jump model training method specific embodiment,
In the embodiment, each sample set that can be concentrated according to training sample carries out the page jump model of initial construction
Training, obtains each page and jumps submodel, constitute final page jump model.So that embodiment of the method two provided
The training method of page jump model not only there is embodiment of the method one to provide possessed by the training method of page jump model
Advantage, additionally it is possible to the obtaining page jump model according to the training of all pages of target APP of the task is split as multiple subtasks,
So as to improve the training effectiveness of page jump model, especially when the page total quantity of target APP is larger, can effectively mention
The training effectiveness of high page jump model.
In the training method for the page jump model that the embodiment of the method one and embodiment of the method two of above-mentioned offer provide,
Can be according to the training sample set of building, training obtains page jump model, so that page jump model is for providing covering mesh
A series of the of the wholly or largely page of mark APP jump movement.
In addition, the training method of any page jump model based on above-mentioned offer, the embodiment of the present application also provides
A kind of page jump test method, is explained and illustrated below in conjunction with attached drawing.
Embodiment of the method three
Referring to Fig. 7, which is the flow chart for the page jump test method that the application embodiment of the method three provides.
Page jump test method provided by the embodiments of the present application, comprising:
S701: using page jump model trained in advance, acting according to the page operation that page jump model provides, right
Target APP carries out page jump test.
Wherein, page jump model is any page jump provided using embodiment of the method one or embodiment of the method two
The page jump model that the training method training of model obtains.
The above are a kind of embodiments for the page jump test method that the application embodiment of the method three provides, in the implementation
In mode, it can use page jump model and page jump test carried out to target APP.Since page jump model is use side
The page jump that the training method training for any page jump model that method embodiment one or embodiment of the method two provide obtains
Model, thus, page jump test method can be jumped between all pages or most of page of target APP, and
All pages or most of page to target APP traverse, to improve the page coverage to target APP.
Further, since trained page jump model is that a deep learning method introduced is instructed based on the above embodiment in advance
It gets, so, the page operation that page jump model provides acts (that is, a series of jump movement), is not only able to guarantee
Biggish page coverage, additionally it is possible to guarantee shorter traverse path, when so that carrying out page jump test to target APP, energy
It is enough to carry out page jump according to the shorter traverse path, reduce invalid page jump movement and duplicate page jump
The generation of movement, to improve the traversal speed to target APP.
In addition, present invention also provides page jump test sides in order to improve the control visiting coverage on each page
The another embodiment of method, in this embodiment, this method further include:
S702: it before jumping to next page from current page using page jump model, is grasped using the preset page
Make tool, in the page operation behavior with randomness of current page simulation real user, and to the behaviour of page operation tool
Make time progress timing;After timing reaches the first preset duration, next page is jumped to.
Page operation tool can be it is any can be to the tool that target APP is tested.
As an example, page operation tool can be monkey tool, wherein monkey tool can simulate randomness compared with
High manual operation realizes that random point is carried out on current page hits movement;Moreover, the random click action may trigger and work as
Control on the preceding page, it is also possible to not trigger the control on current page.In this way, not only increasing the control triggering of target APP
Randomness, also enable page jump test method more accurately to simulate the higher manual operation of randomness.
It should be noted that " click " is for indicating the action triggers that monkey tool uses the page in above-mentioned example
Behavior, still, in this application, monkey tool not only can be using the triggering behavior of " click " or using " sliding "
Triggering behavior, this can also be not especially limited using other triggering behaviors, the application.
First preset duration can be preset, for example, the first preset duration can be set according to application scenarios.
It should be noted that do not have between S701 and S702 it is fixed execute sequence, S701 and S702 can be executed sequentially,
Also S702 and S701 can be executed sequentially, may also be performed simultaneously S701 and S702.
The above are the another embodiments of page jump test method provided by the embodiments of the present application, preferably
In, it can use page jump model and page jump test carried out to target APP, and worked as using preset page operation tool
The page operation behavior with randomness of real user is simulated on the preceding page.In this way, page jump test method can not only
It is jumped between all pages of target APP, so that all pages to target APP traverse, to improve pair
The traversal speed and page coverage of target APP;Moreover, because preset page operation tool can simulate the band of real user
There is the page operation behavior of randomness, page operation tool is randomly triggered and is not related in some page jump models
And control, thus, which can not only test the control being related in page jump model,
Can also in page jump model without reference to control test, so as to improve the control visiting on each page covering
Degree, to improve the control visiting coverage of target APP.
Training method based on any page jump model that above method embodiment one and embodiment of the method two provide
Embodiment carried out below in conjunction with attached drawing the embodiment of the present application also provides a kind of training device of page jump model
Explanation and illustration.
Installation practice one
A kind of training device of page jump model will be introduced in the present embodiment, and related content refers to the above method
Embodiment one and embodiment of the method two.
Referring to Fig. 8, which is that the structure of the training device for the page jump model that the application Installation practice one provides is shown
It is intended to.
The training device 800 of page jump model provided by the embodiments of the present application, comprising:
Construction unit 801, for constructing training sample set;The training sample set includes covering the target APP
Each sample data of all pages, the sample data include the sample action executed on target APP, execute the sample
Initial page mark before movement executes target pages mark after the sample action;
Training unit 802, for according to the training sample set, training to obtain page jump model, the page jump
Model is used to provide a series of the of the wholly or largely page for covering the target APP and jumps movement.
In a kind of implementation of the present embodiment, the construction unit 801 includes:
Subelement is generated, for generating the relation data that jumps of target APP, the relation data that jumps includes the target
Relationship is jumped between each page of APP;
Subelement is traversed, for the page jump model using initial construction, relation data is jumped according to described in and provides respectively
A sample action, and by executing each sample action to traverse all pages of the target APP;
Subelement is constructed, for constructing the training sample set according to the page traverse path to the target APP.
In a kind of implementation of the present embodiment, the generation subelement, comprising:
First trigger module realizes the page for each effective control by the homepage for triggering the target APP
It jumps, to enter under the homepage each first from the page;
Second trigger module, for realizing page jump from each effective control of the page by triggering described first,
To enter described first from each second under the page from the page, until traversal has jumped all pages of the target APP;
Submodule is generated, for according to each page for jumping correspondence, generate the target APP to jump relationship number
According to.
In a kind of implementation of the present embodiment, the training unit 802 is specifically used for:
According to the training sample set, the page jump model of initial construction is trained, the final page is obtained and jumps
Revolving die type;
Alternatively, being carried out according to each sample set that the training sample is concentrated to the page jump model of initial construction
Training, obtains each page and jumps submodel, constitute final page jump model.
In a kind of implementation of the present embodiment, the training unit 802, comprising:
Training subelement is criticized, T batch training is carried out for the page jump model to initial construction, trains for described batch and uses
In carrying out parameter update to current page jump model using K sample data, T is greater than or equal to 1;
Subelement is recycled, for after criticizing training, continuing batch training to the progress of page jump model T times at T times,
Until page jump model meets training termination condition.
In a kind of implementation of the present embodiment, the sample data further include: executing the sample action is later
The no traversal that the page occurs updates.
In a kind of implementation of the present embodiment, the sample data further include: the corresponding income of the sample action
Value;Wherein,
It is updated if executing and the traversal of the page occurring after the sample action, the corresponding financial value of the sample action is
First financial value;
If there is no before and after the update of the traversal of the page and the execution sample action after executing the sample action
The page is different, then the corresponding financial value of the sample action is the second financial value;
If there is no before and after the update of the traversal of the page and the execution sample action after executing the sample action
The page is identical, then the corresponding financial value of the sample action is third financial value;
First financial value, second financial value, the third financial value size successively successively decrease.
In a kind of implementation of the present embodiment, described batch of trained subelement, comprising:
Generation module, for generating the corresponding a reference value of K sample data using current baseline network model, and
The corresponding predicted value of K sample data is generated using current page jump model;
Update module, for carrying out parameter update to current page jump model according to K a reference value and predicted value,
Complete primary described batch of training;
Wherein, the model parameter of the page jump model obtained after T described batch of training is get copied to current reference net
Network model;The a reference value is to hold after executing the sample action in corresponding sample data and entering target pages by initial page
The maximum value in each total revenue that each different series movement of row generates;The predicted value be since the initial page it
The total revenue that prediction action afterwards generates.
Further, the embodiment of the present application also provides a kind of training equipment of page jump model, comprising: processor,
Memory, system bus;
The processor and the memory are connected by the system bus;
The memory includes instruction, described instruction for storing one or more programs, one or more of programs
The processor is set to execute any embodiment party of the training method of above-mentioned page jump model when being executed by the processor
Formula.
Further, described computer-readable to deposit the embodiment of the present application also provides a kind of computer readable storage medium
Instruction is stored in storage media, when described instruction is run on the terminal device, so that the terminal device executes the above-mentioned page
Jump any embodiment of the training method of model.
Further, the embodiment of the present application also provides a kind of computer program product, the computer program product exists
When being run on terminal device, so that the terminal device executes any embodiment party of the training method of above-mentioned page jump model
Formula.
Any embodiment based on the page jump test method that above method embodiment three provides, the application are implemented
Example additionally provides a kind of page jump test device, is explained and illustrated below in conjunction with attached drawing.
Installation practice two
A kind of page jump test device will be introduced in the present embodiment, and related content refers to above method embodiment
Three.
Referring to Fig. 9, which is the structural schematic diagram for the page jump test device that the application Installation practice two provides.
Page jump test device 900 provided by the embodiments of the present application, comprising:
Test cell 901, for being provided according to the page jump model using page jump model trained in advance
Page operation movement carries out page jump test to target APP;
Wherein, the page jump model is that the page that the device training provided using above-mentioned apparatus embodiment one is obtained is jumped
Any embodiment of revolving die type.
In a kind of implementation of the present embodiment, described device further include:
Analogue unit, for utilizing before jumping to next page from current page using the page jump model
Preset page operation tool simulates the page operation behavior with randomness of real user in the current page, and right
The operating time of the page operation tool carries out timing;
Jump-transfer unit, for jumping to the next page after timing reaches the first preset duration.
Further, the embodiment of the present application also provides a kind of page jump test equipments, comprising: processor, memory,
System bus;
The processor and the memory are connected by the system bus;
The memory includes instruction, described instruction for storing one or more programs, one or more of programs
The processor is set to execute any embodiment of above-mentioned page jump test method when being executed by the processor.
Further, described computer-readable to deposit the embodiment of the present application also provides a kind of computer readable storage medium
Instruction is stored in storage media, when described instruction is run on the terminal device, so that the terminal device executes the above-mentioned page
Jump any embodiment of test method.
Further, the embodiment of the present application also provides a kind of computer program product, the computer program product exists
When being run on terminal device, so that the terminal device executes any embodiment of above-mentioned page jump test method.
As seen through the above description of the embodiments, those skilled in the art can be understood that above-mentioned implementation
All or part of the steps in example method can be realized by means of software and necessary general hardware platform.Based on such
Understand, substantially the part that contributes to existing technology can be in the form of software products in other words for the technical solution of the application
It embodies, which can store in storage medium, such as ROM/RAM, magnetic disk, CD, including several
Instruction is used so that a computer equipment (can be the network communications such as personal computer, server, or Media Gateway
Equipment, etc.) execute method described in certain parts of each embodiment of the application or embodiment.
It should be noted that each embodiment in this specification is described in a progressive manner, each embodiment emphasis is said
Bright is the difference from other embodiments, and the same or similar parts in each embodiment may refer to each other.For reality
For applying device disclosed in example, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place
Referring to method part illustration.
It should also be noted that, herein, relational terms such as first and second and the like are used merely to one
Entity or operation are distinguished with another entity or operation, without necessarily requiring or implying between these entities or operation
There are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant are intended to contain
Lid non-exclusive inclusion, so that the process, method, article or equipment including a series of elements is not only wanted including those
Element, but also including other elements that are not explicitly listed, or further include for this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that
There is also other identical elements in process, method, article or equipment including the element.
The foregoing description of the disclosed embodiments makes professional and technical personnel in the field can be realized or use the application.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the application.Therefore, the application
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (17)
1. a kind of training method of page jump model characterized by comprising
Construct training sample set;The training sample set includes each sample number for covering all pages of the target APP
Include the sample action executed on target APP according to, the sample data, execute initial page mark before the sample action
Know, execute target pages mark after the sample action;
According to the training sample set, training obtains page jump model, and the page jump model is for providing described in covering
The a series of of the wholly or largely page of target APP jump movement.
2. the method according to claim 1, wherein the building training sample set includes:
The relation data that jumps of target APP is generated, it is described to jump between each page that relation data includes the target APP
Relationship can be jumped;
Using the page jump model of initial construction, relation data is jumped according to described in and provides each sample action, and by holding
Each sample action go to traverse all pages of the target APP;
According to the page traverse path to the target APP, the training sample set is constructed.
3. according to the method described in claim 2, it is characterized in that, the generation target APP's jumps relation data, comprising:
Page jump is realized by each effective control of the homepage of the triggering target APP, to enter the homepage
Under each first from the page;
Page jump is realized from each effective control of the page by triggering described first, to enter described first under the page
Each second from the page, until traversal has jumped all pages of the target APP;
According to each page for jumping correspondence, generate the target APP jumps relation data.
4. method according to any one of claims 1 to 3, which is characterized in that described according to the training sample set, training
Obtaining page jump model includes:
According to the training sample set, the page jump model of initial construction is trained, obtains final page jump mould
Type;
Alternatively, the page jump model of initial construction is trained according to each sample set that the training sample is concentrated,
It obtains each page and jumps submodel, constitute final page jump model.
5. according to the method described in claim 4, it is characterized in that, the page jump model to initial construction is instructed
Practice, comprising:
Carry out T batch training to the page jump model of initial construction, described batch of training is for using K sample data to working as
Preceding page jump model carries out parameter update, and T is greater than or equal to 1;
At T times after criticizing training, continue to carry out page jump model T batch training, up to page jump model is full
Until foot training termination condition.
6. according to the method described in claim 5, it is characterized in that, the sample data further include: execute the sample action
The traversal that the page whether occurs later updates.
7. according to the method described in claim 6, it is characterized in that, the sample data further include: the sample action is corresponding
Financial value;Wherein,
It is updated if executing and the traversal of the page occurring after the sample action, the corresponding financial value of the sample action is first
Financial value;
If the traversal after executing the sample action there is no the page updates and executes the page before and after the sample action
Difference, then the corresponding financial value of the sample action is the second financial value;
If the traversal after executing the sample action there is no the page updates and executes the page before and after the sample action
Identical, then the corresponding financial value of the sample action is third financial value;
First financial value, second financial value, the third financial value size successively successively decrease.
8. the method according to the description of claim 7 is characterized in that it is described using K sample data to current page jump
Model carries out parameter update, comprising:
The corresponding a reference value of K sample data is generated using current baseline network model, and is jumped using the current page
Revolving die type generates the corresponding predicted value of K sample data;
According to K a reference value and predicted value, parameter update is carried out to current page jump model, completes primary described batch of instruction
Practice;
Wherein, the model parameter of the page jump model obtained after T described batch of training is get copied to current baseline network mould
Type;The a reference value is to execute respectively after executing the sample action in corresponding sample data and entering target pages by initial page
The maximum value in each total revenue that a different series movement generates;The predicted value be since the initial page after
Predict the total revenue that action generates.
9. a kind of page jump test method characterized by comprising
Using page jump model trained in advance, acted according to the page operation that the page jump model provides, to target
APP carries out page jump test;
Wherein, the page jump model is the page jump obtained using the described in any item method training of claim 1 to 8
Model.
10. according to the method described in claim 9, it is characterized in that, the method also includes:
Before jumping to next page from current page using the page jump model, preset page operation work is utilized
Tool, in the page operation behavior with randomness of current page simulation real user, and to the page operation tool
Operating time carry out timing;
After timing reaches the first preset duration, the next page is jumped to.
11. a kind of training device of page jump model characterized by comprising
Construction unit, for constructing training sample set;The training sample set includes all pages for covering the target APP
Each sample data, before the sample data includes the sample action executed on target APP, executes the sample action
Initial page mark, execute target pages mark after the sample action;
Training unit, for according to the training sample set, training to obtain page jump model, and the page jump model is used for
It provides and covers a series of the of the wholly or largely page of the target APP and jump movement.
12. device according to claim 11, which is characterized in that the construction unit includes:
Subelement is generated, for generating the relation data that jumps of target APP, the relation data that jumps includes the target APP
Each page between jump relationship;
Subelement is traversed, for the page jump model using initial construction, relation data is jumped according to described in and provides each sample
This movement, and by executing each sample action to traverse all pages of the target APP;
Subelement is constructed, for constructing the training sample set according to the page traverse path to the target APP.
13. device according to claim 11 or 12, which is characterized in that the training unit is specifically used for:
According to the training sample set, the page jump model of initial construction is trained, obtains final page jump mould
Type;
Alternatively, the page jump model of initial construction is trained according to each sample set that the training sample is concentrated,
It obtains each page and jumps submodel, constitute final page jump model.
14. device according to claim 13, which is characterized in that the training unit, comprising:
Training subelement is criticized, T batch training is carried out for the page jump model to initial construction, described batch of training is for making
Parameter update is carried out to current page jump model with K sample data, T is greater than or equal to 1;
Subelement is recycled, for after criticizing training, continuing batch training to the progress of page jump model T times at T times, up to
Until page jump model meets training termination condition.
15. device according to claim 14, which is characterized in that the sample data further include: it is dynamic to execute the sample
The traversal for making whether to occur later the page updates.
16. a kind of page jump test device characterized by comprising
Test cell, for being grasped according to the page that the page jump model provides using page jump model trained in advance
It acts, page jump test is carried out to target APP;
Wherein, the page jump model is that the page obtained using the described in any item device training of claim 11 to 15 is jumped
Revolving die type.
17. device according to claim 16, which is characterized in that described device further include:
Analogue unit, for before jumping to next page from current page using the page jump model, utilization to be preset
Page operation tool, in the page operation behavior with randomness of current page simulation real user, and to described
The operating time of page operation tool carries out timing;
Jump-transfer unit, for jumping to the next page after timing reaches the first preset duration.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910281458.1A CN109977029A (en) | 2019-04-09 | 2019-04-09 | A kind of training method and device of page jump model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910281458.1A CN109977029A (en) | 2019-04-09 | 2019-04-09 | A kind of training method and device of page jump model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109977029A true CN109977029A (en) | 2019-07-05 |
Family
ID=67083701
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910281458.1A Pending CN109977029A (en) | 2019-04-09 | 2019-04-09 | A kind of training method and device of page jump model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109977029A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110781092A (en) * | 2019-11-07 | 2020-02-11 | 北京博睿宏远数据科技股份有限公司 | Script testing method, device, terminal and storage medium |
CN110990258A (en) * | 2019-10-29 | 2020-04-10 | 贝壳技术有限公司 | Android system-based APP testing method and device, readable storage medium and processor |
CN111694756A (en) * | 2020-07-31 | 2020-09-22 | 北京字节跳动网络技术有限公司 | Application program testing method and device |
CN111694755A (en) * | 2020-07-31 | 2020-09-22 | 北京字节跳动网络技术有限公司 | Application program testing method and device, electronic equipment and medium |
CN111694753A (en) * | 2020-07-30 | 2020-09-22 | 北京字节跳动网络技术有限公司 | Application program testing method and device and computer storage medium |
CN111857719A (en) * | 2020-07-31 | 2020-10-30 | 北京字节跳动网络技术有限公司 | Data processing method, device, electronic equipment and medium |
CN112699046A (en) * | 2021-01-12 | 2021-04-23 | 北京字节跳动网络技术有限公司 | Application program testing method and device, electronic equipment and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080154663A1 (en) * | 2006-12-20 | 2008-06-26 | Sun Microsystems, Inc. | Execution of a BPEL process by simulating partners |
CN106649122A (en) * | 2016-12-28 | 2017-05-10 | Tcl集团股份有限公司 | Model construction method and device for terminal application |
CN107329893A (en) * | 2017-06-21 | 2017-11-07 | 广州酷狗计算机科技有限公司 | Traversal method, device and the storage medium of application interface |
CN108304324A (en) * | 2018-01-22 | 2018-07-20 | 百度在线网络技术(北京)有限公司 | Method for generating test case, device, equipment and storage medium |
CN108334439A (en) * | 2018-03-14 | 2018-07-27 | 百度在线网络技术(北京)有限公司 | A kind of method for testing pressure, device, equipment and storage medium |
-
2019
- 2019-04-09 CN CN201910281458.1A patent/CN109977029A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080154663A1 (en) * | 2006-12-20 | 2008-06-26 | Sun Microsystems, Inc. | Execution of a BPEL process by simulating partners |
CN106649122A (en) * | 2016-12-28 | 2017-05-10 | Tcl集团股份有限公司 | Model construction method and device for terminal application |
CN107329893A (en) * | 2017-06-21 | 2017-11-07 | 广州酷狗计算机科技有限公司 | Traversal method, device and the storage medium of application interface |
CN108304324A (en) * | 2018-01-22 | 2018-07-20 | 百度在线网络技术(北京)有限公司 | Method for generating test case, device, equipment and storage medium |
CN108334439A (en) * | 2018-03-14 | 2018-07-27 | 百度在线网络技术(北京)有限公司 | A kind of method for testing pressure, device, equipment and storage medium |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110990258A (en) * | 2019-10-29 | 2020-04-10 | 贝壳技术有限公司 | Android system-based APP testing method and device, readable storage medium and processor |
CN110781092A (en) * | 2019-11-07 | 2020-02-11 | 北京博睿宏远数据科技股份有限公司 | Script testing method, device, terminal and storage medium |
CN111694753A (en) * | 2020-07-30 | 2020-09-22 | 北京字节跳动网络技术有限公司 | Application program testing method and device and computer storage medium |
CN111694753B (en) * | 2020-07-30 | 2023-04-11 | 北京字节跳动网络技术有限公司 | Application program testing method and device and computer storage medium |
CN111694756A (en) * | 2020-07-31 | 2020-09-22 | 北京字节跳动网络技术有限公司 | Application program testing method and device |
CN111694755A (en) * | 2020-07-31 | 2020-09-22 | 北京字节跳动网络技术有限公司 | Application program testing method and device, electronic equipment and medium |
CN111857719A (en) * | 2020-07-31 | 2020-10-30 | 北京字节跳动网络技术有限公司 | Data processing method, device, electronic equipment and medium |
CN112699046A (en) * | 2021-01-12 | 2021-04-23 | 北京字节跳动网络技术有限公司 | Application program testing method and device, electronic equipment and storage medium |
CN112699046B (en) * | 2021-01-12 | 2024-03-29 | 抖音视界有限公司 | Application program testing method and device, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109977029A (en) | A kind of training method and device of page jump model | |
Openshaw | Some suggestions concerning the development of artificial intelligence tools for spatial modelling and analysis in GIS | |
CN112116092B (en) | Interpretable knowledge level tracking method, system and storage medium | |
CN115659281B (en) | Method and device for fusing adaptive acceleration operators | |
CN111008693B (en) | Network model construction method, system and medium based on data compression | |
CN106326346A (en) | Text classification method and terminal device | |
CN105550746A (en) | Training method and training device of machine learning model | |
CN107544960A (en) | A kind of inference method activated based on Variable-Bindings and relation | |
CN112000772A (en) | Sentence-to-semantic matching method based on semantic feature cube and oriented to intelligent question and answer | |
El Gourari et al. | The implementation of deep reinforcement learning in e-learning and distance learning: Remote practical work | |
US20220383119A1 (en) | Granular neural network architecture search over low-level primitives | |
CN116136870A (en) | Intelligent social conversation method and conversation system based on enhanced entity representation | |
CN113177393B (en) | Method and apparatus for pre-training language model for improved understanding of web page structure | |
Li et al. | A Genetic Causal Explainer for Deep Knowledge Tracing | |
CN111695688B (en) | Model training method, device, equipment and storage medium | |
CN113723566A (en) | Model training method and device | |
He et al. | ADKT: Adaptive deep knowledge tracing | |
Paramita et al. | PARACACHE: Educational Simulator for Cache and Virtual Memory | |
Li | Personalized Dialogue Generation Method of Chat Robot Based on Topic Perception | |
CN117893575B (en) | Ship motion prediction method and system with self-attention mechanism integrated by graph neural network | |
KR102585925B1 (en) | Apparatus for automatically collecting learning data and method therefor | |
Chen et al. | XGBoost-based student performance prediction in tiered instruction | |
CN110059310A (en) | Extending method and device, electronic equipment, the storage medium of hypernym network | |
CN110727768B (en) | Candidate answer sentence generation and natural language selection method and system | |
Ye et al. | A Deep Dive into Keras |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190705 |