Training a Test Agent to Increase Code Coverage Based on DQN for Web Applications
| Autor: | HO, WEI-HANG, 何威杭 |
|---|---|
| Rok vydání: | 2019 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 107 There are many studies that use Web Crawler to automatically explore web applications. However, in order to achieve a higher code coverage, human intervention is often required to rearrange the crawling sequence and provide test inputs for input fields. To achieve full automation, this thesis aims to train a neural-network agent which can manipulate web applications intelligently. In the training process, DQN algorithm of Deep Reinforcement Learning is used. The agent observes the current state of the environment (web application) under explored, and then selects an action to execute. The environment provides the code coverage, which is used as the reward of the action. The training enables the agent to maximize cumulative reward (i.e, code coverage) by learning the manipulation (i.e., selecting the order of actions and the values of the input fields) of the web application. Our experimental results show that a trained agent can indeed select an appropriate action sequence and choose the values for the input fields when exploring a web application, and achieves a higer code coverage in comparison with a traditional web crawler. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|