Research On Task Allocation Technology In CrowdSourcing

In recent years, with the rapid development of Internet and the economic globalization, more and more people, companies, organizations, institutions began to seek ways to solve the problem on the Internet, gradually formed a new business model, called crowdsourcing. Crowdsourcing has emerged as a cheap and efficient method of obtaining solutions for simple tasks that are diffcult for computers to solve but possible for huamans. Since the beginning of 2009, crowdsourcing has received wide attention in various fields, which gradually becomes a new hotspot. In this field, task allocation is a key problem. This thesis mainly studies the task allocation problem.(1) This thesis first introduces the unknown expert crowdsourcing environment. In this environment, workers’ cost and work quality are diverse. Supposing that workers has to submit their cost before they receive tasks, workers’quality of completing the task is unknown and requester has a limited budget. Therefore, it is necessary to use learning method to learn the quality of workers based on the feedback to allocate tasks. The expert crowdsourcing model is similar to the multi-armed bandits problem. In this thesis, we adopt the method of solving multi armed bandits, which is divided into two steps. First, we use part of the total budget to evaluate quality of some workers’ work, and then simulate the function according to the quality results. In this way, we can know the quality of the remaining workers. In addition, we sort the workers according to the quality per unit, using greedy algorithm for task allocation in order to maximize the utility of the requester.(2) The thesis makes a price for task allocation. Under the condition of considering the budget, combining with the gold standard test methods and pricing mechanism design, task allocation problem is further studied. Assuming that each worker submits a bid indicating his/her cost per task and the maximal number of tasks he/she wishes to perform before receiving a task. First, we release some tasks to workers that have the answer based on the results they submit, and estimate the work quality of each worker. Secondly, we make a price according to the bid t submited by wokers. Finally, we sort workers in a descending order by their quality, using the greedy algorithm for task allocation. Combined with actual situation, this model is more suitable for simple tasks whose quality is easily measurable, such as labeling an image containing a face or not, answering a multiple-choice question, or typo correction tasks, etc.(3) Finally, the main components of the platform are analyzed, and the function modules of these components are also introduced. Furthermore, based on the algorithm mentioned in this paper, the task allocation function is preliminary realize by using C# language; according to the current understanding of the knowledge, a simple crowdsourcing platform is established, and the most basic function module of task requester and worker is realized.