KR102292901B1 - Method and apparatus for generating test case for web pages - Google Patents

Abstract

A source code collection module that collects HTML source code from a web page to be tested from a host server, a preprocessing module that preprocesses the collected HTML source code by a low-dimensional vector mapping method, and a convolution neural network (CNN) engine. A classification module that classifies HTML function codes, a test case generation module that loads test case generation data for the function classification, and creates a test case by using case generation data and code level analysis data, and functional elements used for function classification Disclosed is a website testing device including a script generating module that automatically generates a script by using their attribute values.

Description

Method and device for generating test cases for web pages {METHOD AND APPARATUS FOR GENERATING TEST CASE FOR WEB PAGES}

The present invention relates to a functional type classification of a part requiring a functional test in a web page using artificial neural networks, and a method for automatically generating test cases and scripts for each functional type. Through this, the code can be measured through data preprocessing of the low-dimensional vector mapping method, and the function code found in the code of the web page to be tested is classified according to the label of the existing data through a convolution artificial neural network. And, it relates to a method and system for automatically generating a script after generating an appropriate test case through test case generation data for functions belonging to each label. As another embodiment, the unlabeled function code found in the code of the web page to be tested is extracted and classified by object detection through a convolution artificial neural network, and then each function is tested. It relates to a method and a system for automatically generating a script for a suitable test case through case creation data.

Testing of functions included in web page development should be essential to secure the quality of services developed during the development process, thereby increasing end-user satisfaction and reducing direct and indirect costs caused by user's defect discovery after launch. do.

In order to efficiently test the web page, it is possible to identify functional elements that interact with the user's input among various display items in the web page, and verify whether the web page operates as intended for each of these functional elements. A test case should be created by designing an appropriate test method. Examples of such functional elements include a login input field for receiving user's login information, a button input field, a data input field, and the like.

However, due to the lack of time and resources, it is difficult to properly test all the functional elements included in the web page when developing the web page, or to ensure consistency of test results due to the difference in tester competency even after performing the test.

In order to solve this problem, it is necessary to be able to effectively create a large number of appropriate test cases for each function so that coverage of a certain level or more can be consistently secured for different functions.

Most of them derive similar test cases from similar functions, but the time and quality vary depending on the competency of the tester, and even when the same person derives test cases for the same function, several factors (understanding the function, details of the requirements document) quality, etc.) may vary

The test case generation and script generation method for each function in the existing web page is a method in which the tester directly checks the function and creates a test case to check the items described in the functional requirements document. For example, in the Republic of Korea The same method as in Patent Publication No. 10-2010-0069147 (Patent Document 1) has been proposed.

In Patent Document 1, a source information acquisition module that acquires source information corresponding to all objects displayed in a web page through input of a corresponding URL, analyzes the source information and organizes each object independently to position each object A location information analysis module that acquires information, and a test result that derives test result information as a result of scripting the location information by mapping the location information to the corresponding source information according to test conditions of mouse events and keyboard events generated by the user A device comprising a module, and an object display module for displaying an object on a screen based on the location information and the source information, and further displaying the test condition for defining and setting attributes and verification points for the object displayed on the screen do. Through this device, in Patent Document 1, a user-friendly user interface (UI) environment in which dynamic testing is possible by setting various test conditions from a user's point of view rather than a developer's point of view, detecting an object, and analyzing the correlation thereof By providing, it is possible to conduct quality testing on all or individual objects displayed on the website.

Korean Patent Publication No. 10-2010-0069147

However, in the website quality test method including Patent Document 1, although the user can easily input the functional elements to be tested, various functional elements or objects included in the web page that the tester wants to test It does not provide a method for easily identifying and verifying them.

In particular, various functional elements including the login input box, button input box, and data input box in the web page must be verified in different ways according to their types and properties, but how to test each functional element is simply visually inspected. The current method of discriminating and proceeding has a problem that not only the efficiency is lowered, but also the completeness of the test varies depending on the individual tester.

The present invention has been made to solve the above problems, and various functional elements in a web page to be tested are configured by a convolutional neural network (CNN) engine according to attributes such as names, numerical values, and comments associated with each functional element. After classifying into a plurality of categories, collecting the categories with high relevance to the plurality of categories classified in this way, and collecting HTML codes corresponding to functional elements belonging to the corresponding category for each collected category, the tester can be more efficiently This is to enable verification of the corresponding functional elements in the web page.

As an embodiment of the present invention, the CNN engine may include an embedding layer, a convolution layer, a Max-Pooling layer, and a dropout layer.

As an embodiment of the present invention, the CNN engine may use hierarchical clustering analysis.

As an embodiment of the present invention, the CNN engine may use hierarchical cluster analysis. In addition, as an embodiment of the present invention, HTML codes for each function are collected, code patterns of functions found in the test target web page, Based on the properties and characteristics, the function is classified into one of the labels of the data, and test cases are generated using the code level analysis data of the code to be tested based on the test case generation data for the classification to ensure test quality above a certain level at all times. It is intended to provide a method and system that can greatly relieve time and resource constraints by automatically generating scripts by using the attribute values of elements collected during classification afterward. As another embodiment, the unlabeled function code found in the code of the web page to be tested is extracted and classified by object detection through a convolution artificial neural network, and then each function is tested. It is intended to provide a method and system for automatically generating a script for a suitable test case through case creation data.

In the present invention, function classification is implemented using a trained engine by constructing a convolutional artificial neural network by utilizing Python's Tensorflow. TensorFlow is a framework specialized for machine learning, and provides most of the functions required for machine learning as modules. The test case creation process uses the data on the test case creation method for each function generated by the test expert group, and the script automatically creates a Selenium script that performs the functions listed in the test case. Also, hierarchical clustering analysis is used among the hierarchical and non-hierarchical clustering analysis, known as algorithmic analysis methods of collected data.

According to the present invention, by collecting the HTML codes corresponding to the functional elements belonging to the corresponding category for each collected category together, the tester can more efficiently verify the corresponding functional elements in the web page, so that the test of the web page efficiency can be increased. In addition, since the test can be conducted in the same or similar manner for each collected category, it can be expected that a certain level of testing will be performed regardless of the experience of the tester.

1 is a view exemplarily showing the configuration of an apparatus for testing the quality of a website according to an embodiment of the present invention;
FIG. 2 is a diagram showing a method of testing the quality of a target website using the website quality testing apparatus shown in FIG. 1; FIG.
3 is a diagram showing the conversion of the code into a sequence of each word ID through the dictionary;
4 is a diagram illustrating an example of generating a dendrogram according to an embodiment of the present invention.
5 is a diagram illustrating an example of a web page for testing quality according to an embodiment of the present invention.
6 shows an example of an HTML script for implementing the web page shown in FIG. 5 on a computer display.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily carry out the present invention. However, the present invention may be embodied in many different forms and is not limited to the embodiments described herein. Throughout the specification, like reference numerals are assigned to similar parts.

1 is a diagram exemplarily showing the configuration of an apparatus for testing the quality of a website according to an embodiment of the present invention.

1, the test apparatus 100 of the present invention includes a source code collection module 110, a preprocessing module 120, a CNN engine module 130, and a test case generation module ( 140 ), and a script generation module 150 . Hereinafter, each module will be described in detail.

FIG. 2 is a diagram illustrating a method of testing the quality of a target website using the website quality testing apparatus shown in FIG. 1 .

First, the source code collection module 110 collects HTML source codes labeled for each functional element in the web page to be tested from the host server (S210). Next, the pre-processing module 120 pre-processes the collected HTML source code by a low-dimensional vector mapping method (S220). Next, through the already learned CNN deep learning engine (S230), the HTML function code is classified (S240). Then, the test case generation data for the corresponding function classification is loaded (S250), and a test case is generated using the case generation data and code level analysis data (S260). After that, a script is automatically generated by using the attribute values of the function elements used in the function classification (S270).

Function classification is implemented using the trained engine by constructing a convolutional artificial neural network using Python's Tensorflow. TensorFlow is a framework specialized for machine learning, and provides most of the functions required for machine learning as modules. The test case creation process uses the data on the test case creation method for each function generated by the test expert group, and the script automatically creates a Selenium script that performs the functions listed in the test case. Also, hierarchical clustering analysis is used among the hierarchical and non-hierarchical clustering analysis, known as algorithmic analysis methods of collected data.

Hereinafter, each step will be described in detail.

Gather HTML source code labeled by function from host server

First, the source code collection module 110, when a URL is input by a user to receive a web page to be tested from the host server, HTML source code data corresponding to all functional elements displayed in the website from the host server is collected and delivered to the pre-processing module 120 . In this case, it is assumed that the HTML source code data collected from the host server is labeled for each function. The source code data includes not only tag information including embedded information such as FLASH information coded according to a separate form of image (video) in the website, but also tag information to facilitate smooth interworking between tag information and system programs. It can mean a broad concept including JavaScript information.

HTML code data preprocessing with low-dimensional vector mapping method

The preprocessing module 120 receives HTML source code data from the source code collection module 110 and performs preprocessing on it. Specifically, the pre-processing module 120 assigns a unique ID to the tags, attributes, and expected attribute values (words or alphabet combinations meaning specific functions) included in the HTML-delivered source code data, and creates the dictionary after generating tags and attributes. , to convert the attribute value to a number.

3 is a diagram illustrating a code converted into a sequence of word IDs through a dictionary.

Specifically, the code through the dictionary is converted into a sequence of each word ID, as shown in FIG. 3 . Then, a lookup table for vector mapping for each ID is created, each ID is converted into a corresponding vector, and the converted data is placed in the existing order (see FIG. 3 ). Through this pre-processing process, tags, attributes, and attribute values in HTML code can be converted into a list of vectors. The pre-processing module 120 transmits the source code data that has been pre-processed in this way to the CNN engine module 130 so that the CNN engine module 130 performs learning based on the source code data.

clustering

The CNN deep learning engine 130 performs clustering using a hierarchical clustering algorithm on the HTML source code data that has been pre-processed and received from the pre-processing module 120 . That is, clustering is performed according to the degree of similarity of the table obtained through the previous preprocessing, and an algorithm that performs clustering by sequentially and hierarchically integrating individual objects with similar objects or groups using a hierarchical tree model can be utilized. As an example, using a dendrogram, which is a tree-shaped structure indicating the order in which objects are combined, creates a dendrogram as shown in FIG. be able to share After classification, an index for each classification is added to the data to create a label for the data.

In the present invention, as described above, clustering is performed based on hierarchical clustering analysis, but it is also possible to use non-hierarchical clustering analysis.

CNN Deep Learning Engine Module

Various methods of configuring the Convolutional Neural Network (CNN) engine module 130 are known. In an embodiment of the present invention, the CNN engine module 130 has been described as having layers of an embedding layer, a convolution layer, a Max-Pooling layer, and a Dropout layer, but the specific structure of the CNN engine module 130 is not limited thereto. A number of different embodiments may be used.

Embedding Layer

The embedding layer of the CNN engine module 130 maps the word index obtained through the preprocessing process by the preprocessing module 120 to a low-dimensional vector representation.

Convolution Layer

The convolution layer of the CNN engine module 130 is a layer that extracts the features of the input data. From the input data through a filter having several learned w (weight) values, it is possible to help classify the input data. Working to extract features.

First, the data converted into a sequence of vectors is entered as an input to the CNN in batch units. Accordingly, the structure of the input value becomes a four-dimensional tensor [size of batch, code length, number of vector dimensions, number of channels (Depth)]. Batch means the number of codes to be processed at one time, and the number of channels means the number of features in the data. And the structure of the filter is [the height of the filter (number of words), the width of the filter (the number of vector dimensions), the number of channels, the number of filters per filter size]. To apply a filter on a word-by-word basis, the width of the filter can be defined as the number of vector dimensions, and in this case, the size of the filter determines how many words each filter will contain. Since the filter convolutions with the input data, the number of dimensions of the filtered value becomes [batch size, code length - filter size+1, 1, number of filters]. Finally, the nonlinearity can be obtained by using the Recified Linear Unit (ReLU) function.

Max-Pooling Layer :

The Max-pooling layer is a method of extracting the largest value from the activation map after cutting it to an appropriate size. Through the Max-pooling layer, the most essential part is extracted from the state in which the data features are arranged through convolution. That is, the largest value is extracted from the corresponding filter size. The size of the max-pooling area is defined as [number of batch data, code length - filter size + 1, number of vertical dimensions, number of output results]. For example, when Max-pooling 1 batch data by 1 column, the number of dimensions of the value to which Max-pooling is applied becomes [batch size, 1, 1, number of filters].

Dropout Layer :

As a representative method of preventing overfitting of artificial neural networks, it is possible to probabilistically ‘deactivate’ a part of neurons. This prevents simultaneous adaptation of neurons and forces them to learn features individually.

Scores and Predictions :

After going through all the previous steps, the code is classified with the highest score.

Loss and Accuracy :

Using the score obtained through the previous process, we define the loss to be used for learning. The loss means the difference between the prediction of the artificial neural network and the correct answer. Here, the vector values mapped to words along with the filter weights are also updated.

Instantiating and Minimizing the Loss :

Optimize the loss function using the optimizer of TensorFlow.

Classification of HTML function codes through learned engine

Through the engine implemented by utilizing the modules of TensorFlow described above, the part meaning the function in the test target HTML code is identified, and the function of the corresponding part is classified through the learned CNN engine.

After classification, load test case generation data for the corresponding functional classification

Loads data for generating test cases for each classified function. Test cases according to the test design technique applicable to each function and exceptions according to the characteristics of each function are included.

Object detection algorithm

As described above, the embodiment in which the process is carried out in the order of labeled source code collection, preprocessing, clustering, classification index, and test generation data loading for each function has been described. do.

(i) unlabeled source code collection

(ii) All HTML source code preprocessing

(iii) Extraction and classification of functional elements by object detection

(iv) Loading test-generated data for each function

In this way, when functional elements are extracted and classified using object detection, there is an advantage in that labeling and clustering are not required.

Create test cases using case creation data and code-level analysis data

Generates multiple test cases for each function based on the imported data. Through this, it is possible to create test cases that consistently secure coverage above a certain level for all functions.

Scripts are automatically generated by using the attribute values of the elements used to classify functions.

Selenium script is automatically generated by using the attribute values of each element checked during code analysis.

5 is a diagram illustrating an example of a web page for testing quality according to an embodiment of the present invention. The web page of FIG. 5 includes a plurality of dynamic web page components 510 , 520 , 530 , and 540 , and these are components to be subjected to a quality test.

6 shows an example of an HTML script for implementing the web page shown in FIG. 5 on a computer display. In the figure, it is assumed that script portions 610, 620, 630, and 640 are scripts for implementing a plurality of dynamic webpage components 510, 520, 530, 540 shown in FIG. 5, respectively.

The apparatus for testing the quality of a website according to an embodiment of the present invention receives an HTML script as shown in FIG. 6 and collects HTML source codes labeled for each functional element (S210). Next, code data is pre-processed by a low-dimensional vector mapping method for each of the script components, for example, the script parts 610, 620, 630, and 640 shown in FIG.

Next, processing by the CNN deep learning engine is performed (S230), and the HTML code is classified (S240). Next, the test case generation data for the corresponding function classification is loaded (S250), and a test case is generated using the corresponding case generation data and code level analysis data (S260). Next, a script is automatically generated by using the attribute value of the function element used in the function classification (S270).

Although the preferred embodiment of the present invention has been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concept of the present invention as defined in the following claims are also provided. is within the scope of the

110: source code collection module 120: pre-processing module
130: CNN deep learning engine 140: test case generation module
150: script generation module

Claims (8)

A source code collection module that collects HTML source code from the web page to be tested from the host server;
A preprocessing module that preprocesses the collected HTML source code by a low-dimensional vector mapping method;
A classification module for classifying HTML function codes by a convolution neural network (CNN) engine by using a hierarchical cluster analysis;
A test case generation module that loads the test case generation data for the corresponding function classification and creates a test case by using the case generation data and code level analysis data;
Including a script generation module that automatically generates a script based on the attribute values of the function elements used when classifying the function,
Website testing device. The CNN engine of claim 1, wherein the CNN engine comprises layers of an embedding layer, a convolution layer, a Max-Pooling layer, and a Dropout layer.
Website testing device. delete The method of claim 1, wherein the CNN engine uses object detection.
Website testing device. A source code collection step of collecting HTML source code from the web page to be tested from the host server;
A preprocessing step of preprocessing the collected HTML source code by a low-dimensional vector mapping method;
A classification step of classifying HTML function codes by a convolution neural network (CNN) engine by using a hierarchical cluster analysis;
A test case creation step of loading the test case creation data, creating a test case by using the case creation data and code-level analysis data;
Including a script generation step of automatically generating a script based on the attribute values of the function elements used when classifying the function,
How to test your website.
[6] The method of claim 5, wherein the HTML classification step is performed through layers of an embedding layer, a convolution layer, a Max-Pooling layer, and a Dropout layer.
delete 6. The method of claim 5, wherein the CNN engine uses object detection.
How to test your website.

Images