JP6280092B2 - Estimation apparatus and estimation method - Google Patents

Description

  The present invention relates to an estimation apparatus and an estimation method.

  With the widespread use of the Internet, whether or not a web browser can be used comfortably greatly affects whether the user can use the Internet comfortably.

  For this reason, it is necessary to understand and manage the time it takes for the user to operate the browser when browsing the web page and the browser reads and displays the data (hereinafter referred to as “web page display time”). It is important for management.

  In order to grasp the web page display time, the browser is provided with a mechanism such as NavigationTiming, and a method for directly knowing the display time has been widespread (Non-Patent Document 1, Non-Patent Document 2).

Honda et al., `` Web quality degradation isolation using Navigation Timing API '', Communication Quality Study Group 2014. “Navigation Timing”, [online], [October 26, 2015 search], Internet (URL: http://www.w3.org/TR/navigation-timing/) H. Drucker, et al., `` Support Vector Regression Machines '', Advances in Neural Information Processing Systems 9, NIPS 1996,

  However, the NavigationTiming technology is assumed to be used by content providers (web servers) and users (client terminals), and for data observed on the network managed by the carrier in charge of data transfer It is difficult to apply.

  The present invention has been made in view of the above points, and an object thereof is to make it possible to estimate the comfort of a web browser from observation data on a network.

  Therefore, in order to solve the above-described problem, the estimation apparatus performs the determination within each trial for each of a predetermined number of GET requests in the order of early occurrence in each of a plurality of trials related to display on a web browser for a certain web page. Transfer of at least the data necessary for displaying the certain web page to the web browser is completed after the relative occurrence time and the display of the certain web page measured for each trial are instructed. An estimated model generation unit that learns a required time until a predetermined estimated model, and the relative estimated time of each GET request regarding the certain web page observed on the network, the predetermined estimated model And an estimator that estimates the required time for the GET request.

  The comfort of the web browser can be estimated from the observation data on the network.

It is a figure which shows the hardware structural example of the estimation apparatus in embodiment of this invention. It is a figure which shows the function structural example of the estimation apparatus in embodiment of this invention. It is a flowchart for demonstrating an example of the process sequence which an estimation apparatus performs. It is a figure which shows the generation history of HTTP-GET which comprises learning data. It is a figure which shows the example of the log | history of the transfer completion time and display completion time which comprise learning data. It is a figure which shows an example of the extraction result of the record for the learning threshold value about each trial. It is a figure which shows an example of the conversion result to the relative time of the generation time of each HTTP-GET. It is a figure which shows an example of the vector of the transfer completion time or display completion time for every trial.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. First, a process for browsing / displaying a web page will be described. The process is greatly simplified as follows.
(1) A page acquisition request is generated starting from a click by the user.
(2) When the browser requests the transfer of a web page with the GET request of the HTTP (HyperText Transfer Protocol) for the html data of the page indicated by the target URL, the html data is transferred to the browser.
(3) Sub-contents (image data, required script, etc.) are shown in the html data, and the browser sequentially decodes them and obtains them with an HTTP GET request (GET with HTTP).
(4) The browser processes and displays the sub-contents acquired sequentially.

  When browsing a general web page, if it is not encrypted with https, it reads the packet exchange on the network (packet capture), thereby obtaining an individual HTTP GET request (hereinafter referred to as “HTTP-GET”). The transmission status and the transmission time can be grasped. Therefore, omitting the display processing in (4), at least the required time from (1) to (3) (≈ transfer of data necessary for display to the browser after the instruction to display the web page is completed Display completion time can be simply estimated. Hereinafter, the “required time from when the web page display is instructed until the transfer of data necessary for display to the browser” is simply referred to as “transfer completion time”. The required time from (1) to (4) is referred to as “display completion time”.

However, there are the following two problems to specify the “transfer completion time” by observing HTTP-GET.
(A) Depending on the creation of the web page, HTTP-GET is continuously generated even after the display of the web page is completed by automatic update or the like.
(B) When the same web page is repeatedly displayed on the same terminal, the number of HTTP-GET generated until the display is completed is not necessarily constant.

  First, according to (a), the “timing at which generation of HTTP-GET ends” for one web page cannot be determined within the scope of observing the generation status of the GET request on the network. Does not correspond to the completion of HTTP-GET transfer.

  Further, according to (b), a predetermined threshold is determined in advance for the number of HTTP-GETs for each web page, the HTTP-GET is transferred when the number of GETs reaches the threshold by observing the HTTP-GET sequence. It is also difficult to judge that the completion is complete.

  Therefore, in the present embodiment, the transfer completion time or the transfer completion time or the time from the occurrence history of HTTP-GET at the stage before the completion of transfer (the completion of (3) above) (the stage until the middle of the above (3)). Estimate the display completion time.

  FIG. 1 is a diagram illustrating a hardware configuration example of an estimation apparatus according to an embodiment of the present invention. The estimation device 10 in FIG. 1 includes a drive device 100, an auxiliary storage device 102, a memory device 103, a CPU 104, an interface device 105, and the like that are mutually connected by a bus B.

  A program that realizes processing in the estimation apparatus 10 is provided by a recording medium 101 such as a CD-ROM. When the recording medium 101 storing the program is set in the drive device 100, the program is installed from the recording medium 101 to the auxiliary storage device 102 via the drive device 100. However, the program need not be installed from the recording medium 101 and may be downloaded from another computer via a network. The auxiliary storage device 102 stores the installed program and also stores necessary files and data.

  The memory device 103 reads the program from the auxiliary storage device 102 and stores it when there is an instruction to start the program. The CPU 104 executes a function related to the estimation device 10 according to a program stored in the memory device 103. The interface device 105 is used as an interface for connecting to a network.

  In addition, the estimation apparatus 10 may be configured by a plurality of computers each having the configuration illustrated in FIG.

  FIG. 2 is a diagram illustrating a functional configuration example of the estimation device according to the embodiment of the present invention. 2, the estimation apparatus 10 includes a learning data acquisition unit 11, a threshold determination unit 12, an estimation model generation unit 13, an estimation unit 14, and the like. Each of these units is realized by processing that one or more programs installed in the estimation apparatus 10 cause the CPU 104 to execute. The estimation device 10 also uses the learning data storage unit 15. The learning data storage unit 15 can be realized by using, for example, a storage device that can be connected to the auxiliary storage device 102 or the estimation device 10 via a network.

  The learning data acquisition unit 11 acquires learning data for estimating a transfer completion time or a display completion time for an evaluation target web page (hereinafter referred to as “evaluation target page”) related to a preset URL. If the estimation target is the transfer completion time, learning data regarding the transfer completion time is acquired, and if the estimation target is the display completion time, learning data regarding the display completion time is acquired. It is set in advance whether the estimation target is the transfer completion time or the display completion time. The learning data is acquired by executing, in parallel, acquisition of network capture data, detection of timing such as completion of data transfer regarding the web page in the web browser or completion of display of the web page. The learning data acquisition unit 11 includes a web page display instruction and display repeatedly performed a plurality of times for the same evaluation target page, a page transfer completion time or a page display completion time, and a transferred HTTP-GET (HTTP HTTP). Learning data indicating the history of occurrence of (GET request) is acquired. The acquired learning data is stored in the learning data storage unit 15.

  The threshold determination unit 12 determines the number (threshold) of HTTP-GET used for estimating the transfer completion time or the display completion time based on the learning data.

  The estimation model generation unit 13 causes the statistical estimation model to learn the learning data within the threshold range determined by the threshold determination unit 12.

  The estimation unit 14 applies the estimation model obtained by learning the learning data to the observation data (GET request occurrence history) observed on the network, and sets it as an estimation target among the transfer completion time and the display completion time. Estimate who has been.

  Hereinafter, the process procedure which the estimation apparatus 10 performs is demonstrated. FIG. 3 is a flowchart for explaining an example of a processing procedure executed by the estimation apparatus.

  In step S <b> 101, the learning data acquisition unit 11 displays an evaluation target page in which the URL is set in advance on a browser in the estimation device 10, an HTTP-GET occurrence history at that time, and an evaluation target page by the browser The actual value of the transfer completion time or display completion time is acquired. The learning data acquisition unit 11 stores the acquired data (learning data) in the learning data storage unit 15. Note that the HTTP-GET occurrence history is acquired from the start of transfer to the transfer completion time. That is, HTTP-GET after the transfer completion time is not included in the occurrence history.

As for the occurrence history of HTTP-GET, for example, after acquiring packet data with a packet capture tool such as tcpdump in the estimation apparatus 10, the number of times and time of occurrence of HTTP-GET with the capture analysis tool such as tshark It may be obtained by converting into For example, if the capture data file name is dump. If it is pcap, the number of HTTP-GET occurrences and the time can be obtained by executing the following command.
% tshark -r dump.pcap -Y http.request.uri -T fields -e frame.time_epoch-e ......
Alternatively, a proxy may be set in the estimation device 10, and a log of HTTP-GET time information may be acquired from a proxy log when a web page is displayed by a browser.

  On the other hand, the transfer completion time and display completion time of the evaluation target page in the browser can be measured and quantified, for example, by an API (Application Program Interface) of NavigationTiming.

  Note that one execution of step S101 is referred to as “trial”. Step S101 is repeated for the number of trials set in advance (for example, 101 times). As a result, learning data constituted by the information shown in FIGS. 4 and 5 is acquired.

  FIG. 4 is a diagram showing an occurrence history of HTTP-GET constituting the learning data. In FIG. 4, the HTTP-GET occurrence history includes a trial number, a GET request, a time, etc. for each HTTP-GET detected in each trial.

  The trial number is a value indicating what number of trial the GET request is detected. The GET request is a character string indicating the content of the GET request. The time is the time when the GET request is generated (detected). Since the number of GET requests can be different for each trial, the number of rows for each trial can be different in FIG.

  FIG. 5 is a diagram illustrating an example of a transfer completion time and display completion time history constituting the learning data. FIG. 5 shows the transfer completion time and display completion time of the evaluation target page in each trial. In addition, when the estimation target is the transfer completion time, only the transfer completion time may be acquired, and when it is the display completion time, the transfer completion time and the display completion time may be acquired.

  In FIG. 4 and FIG. 5, records having a common trial number are learning data relating to the same trial number. For example, in FIG. 5, the HTTP-GET occurrence history regarding trial 2 (trial whose trial number is 2) can be identified by referring to the record whose trial number is 2 in FIG. 4.

  Note that the trial related to the display of the evaluation target page may be executed in a terminal different from the estimation device 10. In this case, it is only necessary to input learning data in step S101.

  Subsequently, the threshold value determination unit 12 determines a threshold value by applying a preset threshold value determination method and parameters to the learning data (step S102).

  First, the threshold value determination unit 12 aggregates the number of records for each trial in the HTTP-GET occurrence history (FIG. 4), and sorts the aggregation results in ascending order. In the example of FIG. 4, the count result of trial 1 is 4, and the count result of trial 2 is 5.

  Subsequently, the threshold determination unit 12 determines a threshold based on a preset threshold determination method and parameters. In the present embodiment, in “distribution lower percentile designation”, a quantile point determined in advance is determined as a threshold value from the smaller sorted number side.

  For example, if the threshold value determination method is “lower 5th percentile of distribution”, a position 5% from the minimum value of the sorting result (that is, the sixth value from the minimum value in the case of 101 trials) is determined as the threshold value. Is done.

  This is because if the minimum value is used and the sample or the like that has ended abnormally is included, the minimum value becomes extremely small, and there is a possibility that an appropriate number of data necessary for estimation cannot be obtained. Therefore, a certain quantile extraction is used.

  The threshold value determined as described above is referred to as “learning threshold value N”.

  Subsequently, for each trial, the estimated model generation unit 13 records, for each trial, N records corresponding to the learning threshold value N from the smallest time value (that is, N records in the order of occurrence time). A record related to the GET request is extracted (step S103). For trials in which the number of records is less than the learning threshold N, no records are extracted.

  FIG. 6 is a diagram illustrating an example of a record extraction result corresponding to the learning threshold for each trial. In FIG. 6, for each trial in which the record is extracted, the trial number and the occurrence time of N HTTP-GETs are shown. FIG. 6 corresponds to an example in which M trials include N or more learning threshold records.

  Subsequently, for each trial in which the record is extracted, the estimation model generation unit 13 converts the occurrence time of each HTTP-GET into a relative value (relative time) from the occurrence time of the first HTTP-GET of the trial. (Step S104). That is, for the GET-X (X = 1 to N) time of each trial shown in FIG. 6, the difference from the GET-1 time (relative occurrence time within the trial) is calculated.

  FIG. 7 is a diagram illustrating an example of a conversion result of the generation time of each HTTP-GET to a relative time. FIG. 7 shows the conversion result to the relative time for the generation time of each HTTP-GET in each trial shown in FIG. Note that the GET-1 time is discarded because the relative time is always 0 for each trial and there is no information amount. As a result, the number of columns in each trial is N-1.

  Subsequently, the estimation model generation unit 13 generates a transfer completion time or display completion time vector for each trial (the number of GETs is N or more) (step S105).

  FIG. 8 is a diagram illustrating an example of a transfer completion time or display completion time vector for each trial. In FIG. 8, (1) is a transfer completion time vector generated based on the learning data shown in FIG. (2) is a display completion time vector generated based on the learning data shown in FIG. If the estimation target is the transfer completion time, only the transfer completion time vector may be generated. If the estimation target is the display completion time, only the display completion time vector may be generated.

  Subsequently, the estimated model generation unit 13 generates an estimated model based on a preset estimation method (step S106). In the present embodiment, an example in which support vector regression (SVR) is set as the estimation method will be described. However, the estimation method applicable to the present embodiment is not limited to SVR, and other methods may be used. For example, multiple regression or the like may be used. Note that the SVR itself is just a known method of machine learning, and in this embodiment, the usage of input data and mapping with output are the points.

Here, the learning data (X) is configured as follows.
X = (X_1, X_2, ..., X_ (N-1))
However,
X_1 = [vector of length M of relative time of GET-2 of each trial]
= [0.032, 0.200, ..., 0.222] (example in Fig. 7)
X_ (N−1) = [0.233, 0.668,..., 1.223]
Further, the learning data (Y) is configured as follows.
If transfer completion time is to be estimated,
Y = [vector of transfer completion time length M] = [2.5, 2.9,..., 4.1]
If the display completion time is to be estimated,
Y = [vector of display completion time length M] = [3.1, 3.6,..., 4.7]
The estimation model generation unit 13 generates an SVR model E and learns the learning data (X) and the learning data (Y) configured as described above. That is, the model E is caused to learn the relationship between X and Y.

For example, according to the usage of SVR (http://scikit-learn.org) in the scikit-learn library having the SVR function, model E is generated and learned according to the following description. be able to.
E = SVR () #Model generation
E.fit (X, Y) #Learning with learning data Subsequently, the estimation unit 14 applies the model E to the observation data on the network, and estimates the transfer completion time or display completion time for the observation data. (Step S107).

  Specifically, the estimation unit 14 acquires a history of occurrence times of HTTP-GET related to evaluation target pages observed in time series on the network. The estimation unit 14 calculates a relative generation time (relative time) from the first HTTP-GET generation time with respect to the generation time of each HTTP-GET. As a result, a relative time vector X_new is obtained.

For example, it is assumed that the history of observed occurrence times of each HTTP-GET is as follows.
GET occurrence time history: (234.333, 2345.999, 2346.500, 2347.000)
In this case, the value of X_new is as follows.
X_new = [0.666, 1.167, ..., 1.667]
The estimation unit 14 estimates X_new using the estimation model E, and obtains an estimated value T_e.
T_e = E. predict (X_new)
The value of T_e estimated in this way (for example, 3.8) is an estimated value of “transfer completion time” or “display completion time” regarding the observation data. That is, if the learning data (Y) is a vector of transfer completion time, an estimated value of transfer completion time is obtained. If the learning data (Y) is a vector of display completion time, an estimated value of display completion time is obtained.

  Note that estimated values of both the transfer completion time and the display completion time may be obtained.

  In addition, an estimation model is generated for each of a plurality of URLs, and transfer completion time or display completion time is estimated using an estimation model corresponding to a URL related to a GET request observed in the network. Also good.

  As described above, according to the present embodiment, the comfort of the web browser can be estimated from the observation data on the network.

  In the present embodiment, the evaluation target web page is an example of a certain web page. The transfer completion time and the display completion time are measured for each trial, and after the display of the certain web page is instructed, at least the transfer of data necessary for displaying the certain web page to the web browser is completed. It is an example of the time required until.

  As mentioned above, although the Example of this invention was explained in full detail, this invention is not limited to such specific embodiment, In the range of the summary of this invention described in the claim, various deformation | transformation・ Change is possible.

DESCRIPTION OF SYMBOLS 10 Estimation apparatus 11 Learning data acquisition part 12 Threshold value determination part 13 Estimation model production | generation part 14 Estimation part 15 Learning data storage part 100 Drive apparatus 101 Recording medium 102 Auxiliary storage apparatus 103 Memory apparatus 104 CPU
105 Interface device B bus

Claims (4)

Relative occurrence time within each trial for each of a predetermined number of GET requests in order of early occurrence in each of a plurality of trials related to display on a web browser for a certain web page, and measurement for each trial. The time required from the time when the display of the certain web page is instructed until the transfer of the data necessary for displaying the certain web page to the web browser is completed according to a predetermined estimation model. An estimated model generator to be trained;
An estimation unit that estimates the time required for the GET request by applying the predetermined estimation model to a relative generation time of each GET request for the certain web page observed on the network;
The estimation apparatus characterized by having. A threshold determination unit that determines a threshold based on the number of GET requests in each of the plurality of trials;
The estimated model generation unit, in each trial in which the number of GET requests is greater than or equal to the threshold, relative occurrence time within the trial for each of the threshold number of GET requests in order of the occurrence time; A time required until the transfer of data necessary for displaying the certain web page to the web browser is completed after the display of the certain web page is instructed, measured for each trial. Let the predetermined estimation model learn,
The estimation apparatus according to claim 1. Relative occurrence time within each trial for each of a predetermined number of GET requests in order of early occurrence in each of a plurality of trials related to display on a web browser for a certain web page, and measurement for each trial. The time required from the time when the display of the certain web page is instructed until the transfer of the data necessary for displaying the certain web page to the web browser is completed according to a predetermined estimation model. An estimation model generation procedure to be trained;
An estimation procedure for estimating the required time for the GET request by applying the predetermined estimation model to a relative occurrence time of each GET request for the certain web page observed on the network;
Is performed by a computer. The computer executes a threshold determination procedure for determining a threshold based on the number of GET requests in each of the plurality of trials;
The estimation model generation procedure includes a relative generation time in each trial for each of the threshold number of GET requests in order of early generation time in each trial in which the number of GET requests is equal to or greater than the threshold. A time required until the transfer of data necessary for displaying the certain web page to the web browser is completed after the display of the certain web page is instructed, measured for each trial. Let the predetermined estimation model learn,
The estimation method according to claim 3.

Images