众包环境下面向部分序属性域的低费用质量可控的Pareto查询机制研究

Crowdsourcing provides a new problem solving scheme which utilizes the intelligence of crowds to solve machine-hard problems by leveraging a large group of people on the web. The Pareto-optimal query, which plays an irreplaceable role in information retrieval and decision making, is an important content in the field of crowdsourcing services. It is a challenging problem to implement Pareto-optimal queries effectively in a crowdsourcing environment. First, low quality answers are easily caused by the free and loose organization model of crowdsourcing and the difficulty to determine preferences over partial orders. Second, Pareto-optimal query requires a large number of crowdsourcing microtasks, thereby easily giving rise to a high monetary cost. To tackle the above challenges, this project provides a further and systematical study on crowdsourced Pareto-optimal query mechanism over partial orders with quality guarantee, from the aspects of new theories, methods, and different application scenarios. Specifically, we first proposed a new quality-controllable microtask model of dynamic population, starting from the three aspects: preference judgement, microtask assignment, and crowdsourcing answer aggregation. The microtask model minimizes the microtask cost with quality guarantee. Next, we proposed microtask reduction principles and rules based on transitivity in theory, and a filtering-verification solution framework in the methodology, for the purpose of pruning microtasks on the levels of both attribute and object, thereby effectively reducing the overall monetary cost. We also study Pareto-optimal queries for different applications, e.g., multiple queries, and queries for data streams. This research will help to improve the basic theory of data queries and information retrieval, and promote the development of crowdsourcing applications.

众包是一种公开面向互联网大众的通过群体智慧求解问题的新型模式。Pareto查询是众包服务重要的内容，在信息检索、决策支持等方面具有不可替代的作用。本项目针对由于众包松散组织模式并且偏好关系不易判断而导致低质量答案、Pareto查询极易导致大量微任务而带来较大费用开销的问题，从理论到方法再到不同应用场景，系统深入地研究众包环境下面向部分序属性域的低费用质量可控的Pareto查询机制。主要包括：提出一种质量可控的动态人数的微任务模型，从偏好关系判断、微任务分配、众包答案整合三个方面着手，在保证众包质量的同时减少单个微任务费用开销；从理论上研究基于传递性理论的微任务消减基本原理，从方法上构建基于过滤-验证框架的查询方案，从属性和对象两个级别进行微任务剪枝，从而降低总费用开销；进一步研究多查询和数据流这两种不同应用场景下的Pareto查询。该项研究有助于完善查询检索基础理论、促进众包应用发展。

众包是一种新型问题求解模式，推动了共享经济的兴起。数据查询是众包应用和服务的基本内容。本项目以基于众包借助人类智慧来实现面向机器难问题的查询处理为研究背景，研究面向部分序属性域的低费用质量可控的Pareto查询机制。根据部分序属性域偏好关系特征以及众包工作模式，重点针对众包查询处理中的质量控制以及费用控制的相关关键技术开展了深入研究。考虑众包质量和众包费用的取决因素，本项目主要从以下四个方面展开研究：1）研究质量可控的动态人数的众包微任务模型，设计了基于滑动评分和梯度划分的微任务模型、置信度感知的众包答案聚合方案，在实现偏好关系定性定量评价的同时最小化工人人数；2）研究众包工人质量评估以及众包任务分配，提出了基于高斯混合模型的工人质量等级式评估方案，设计了基于任务特征的分配算法，通过选择高质量工人参与任务以及制定分配方案来提高众包答案质量；3）研究面向费用控制基于传递性理论的Pareto查询处理，提出基于过滤-验证框架的查询方案，通过减少微任务数目减低费用开销；4）进一步针对不同应用场景,研究多Pareto查询和连接查询并提出具体的查询算法，为提高众包服务质量、促进共享经济的发展提供理论支撑。