基于众包思想的网站无障碍检测系统的任务分配和质量控制

发布时间：2018-06-15 00:36

本文选题：信息无障碍 + 无障碍衡量标准　；参考：《浙江大学》2017年硕士论文

【摘要】：互联网的蓬勃发展,引领了社会信息化的新潮流,创造了人类社会交流的新空间,给人类的生产生活带来了巨大的影响与改变。然而,当今社会仍然有相当一部分特殊人群,如视力残疾、听力残疾人群等,他们无法像健全人一样获取网上信息,享受互联网带来的便利。快速发展的互联网一定程度上加大了残疾人与健全人之间的信息鸿沟。为了使残疾人能更方便的浏览网站,获取网站信息,首先需要加强网站的无障碍建设。尽管网站无障碍建设的工作开展已久,如今的网站无障碍仍然存在很大的问题。为了发掘网站中存在的访问障碍点,需要对网站进行无障碍评估,Bobby等自动检测工具能在一定程度上检测出网站中存在障碍的检测点,但是只通过自动检测工具是远远不够的,较多的无障碍检测点需要人工参与。基于众包思想的网站无障碍检测系统(下文简称无障碍检测系统)用来辅助用户对网站进行无障碍检测,该系统利用互联网上的用户的群体智慧,让更多的人参与到无障碍检测中来。然而,参与检测的志愿者专业背景、能力等相差较大,简单的随机任务分配不足以利用志愿者的特征,分配给他们合适的任务,因此需要寻找一种能利用志愿者和检测任务的特征来进行任务分配的方法。此外,一个志愿者的结果往往不具备可靠性,一条检测任务通常需要由多个志愿者共同完成。如何控制任务结果的质量是一个需要考虑的问题。考虑到随机分配不能将检测任务分配给较为适合的志愿者,因此,本文提出了一种基于遗传算法和置信度的任务分配方法,该方法结合志愿者和检测任务特征来分配检测任务。本文主要提取了志愿者对不同检测规则的正确率、错误率和放弃率特征,通过遗传算法计算各个特征在任务分配时所占的权重,并用置信度来衡量新的任务分配方式与随机任务分配方式的差距。在任务质量控制方面,本文主要列举了在无障碍检测系统中用到的任务质量控制方法,如恶意检测用户过滤等。针对众包任务结果的最终决策方法,本文提出了一种基于加权可信度的的任务质量控制方法。该方法结合志愿者可信度和任务结果可信度的加权可信度,在此加权可信度的基础上计算使结果误差达到最小时应该采取的可信度的值,以此来控制众包任务质量。最后,本文提取无障碍检测系统的数据,对本文提出的方法进行实验验证,证明本文提出的任务分配方法和众包任务结果质量控制方法都比传统的无障碍检测系统中采取的方法要好。
[Abstract]:The vigorous development of the Internet has led the new trend of social information, created a new space for human social exchanges, and brought great influence and changes to the production and life of human beings. However, there are still some special people in our society, such as visual disability, hearing disability and so on. They can not get the online information and enjoy the convenience brought by the Internet just like the healthy people. The rapid development of the Internet has to some extent increased the information gap between the disabled and the healthy. In order to make the disabled more convenient to visit the website, access to information, the first need to strengthen the accessibility of the site. Despite the long history of accessibility, there are still a lot of problems with accessibility. In order to discover the barriers in the website, it is necessary to evaluate the accessibility of the site. Bobby and other automatic detection tools can detect the barriers in the site to a certain extent, but it is not enough to use automatic detection tools. More accessibility testing points require manual participation. The website barrier-free detection system based on crowdsourcing (hereinafter referred to as the barrier-free detection system) is used to assist users in carrying out barrier-free detection of websites. The system exploits the collective wisdom of users on the Internet. Let more people participate in barrier-free testing. However, the professional background and ability of the volunteers involved in the test vary greatly, and the simple random task assignment is not enough to use the characteristics of the volunteers to assign them suitable tasks. Therefore, we need to find a method that can use the characteristics of volunteers and detection tasks to carry out task assignment. In addition, the results of a volunteer are often not reliable, and a test task usually needs to be completed by more than one volunteer. How to control the quality of task results is a problem to be considered. Considering that random assignment can not assign detection tasks to more suitable volunteers, this paper proposes a task assignment method based on genetic algorithm and confidence, which combines volunteers and detection task characteristics to assign detection tasks. In this paper, the correct rate, error rate and abandonment rate feature of different detection rules are extracted, and the weight of each feature in task assignment is calculated by genetic algorithm. The confidence level is used to measure the difference between the new task assignment and the random task assignment. In the aspect of task quality control, this paper mainly enumerates the task quality control methods used in barrier-free detection system, such as malicious detection, user filtering and so on. For the final decision method of crowdsourcing task results, this paper presents a method of task quality control based on weighted credibility. The method combines the weighted credibility of volunteers and the credibility of task results, and calculates the value of credibility that should be adopted when the result error reaches the minimum on the basis of the weighted credibility, so as to control the quality of crowdsourcing tasks. Finally, the data of the barrier-free detection system is extracted, and the proposed method is verified experimentally. It is proved that both the task assignment method and the crowdsourcing task result quality control method proposed in this paper are better than those in the traditional barrier-free detection system.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.092

【相似文献】