III: Small: A Big Data and Machine Learning Approach for Improving the Efficiency of Two-sided Online Labor Markets

在线劳动力市场（OLM），即按需雇用工人和在线交付产品，已成为当今经济中日益重要的组成部分。这些双边市场为工人和雇主提供了访问大量地理上分散的、半匿名的虚拟交易伙伴的机会。数以百万计的美国人在各种OML平台上从事自由职业，他们为美国经济贡献了数千亿美元。为了支持工人和雇主执行日常业务，OLM平台开发了一些功能和工具。示例包括基于关键字的职位搜索、任务匹配和员工推荐。然而，这些工具中的许多工具都有严重的缺点，并会导致重大问题（例如在将任务暴露给工人时存在偏见）。该项目将收集和分析大量的OLM数据，并开发一套新颖的技术解决方案，以提高当今OLM平台的效率。开发的技术方法将为数据挖掘，机器学习和OLM分析领域做出研究贡献。本项目的研究成果将促进对美国OLM的认识和理解。该项目旨在使多个利益相关者受益，并对社会产生潜在影响，例如为OLM平台提供新的工人-任务匹配工具，更新OLM工人的技能，并改善OLM就业。该项目通过课程模块开发，将研究与教育结合起来，让研究生和本科生参与研究，并为本地K-12学生提供研究展示。该项目侧重于以下三个具体目标（SA）：开发新的技术解决方案，用于发现有用的知识，从大量OLM数据中学习特征表示，研究工人之间的过度竞争问题，减轻OLM中任务的不平衡暴露，开发一种新的双边匹配方法来解决这些问题，分析OLM任务和工人之间的关键技能差距，并开发一种新的前瞻性OLM技能推荐解决方案来弥合差距。该项目将开发一种上下文感知的深度广泛方法，用于从大量文本中挖掘OLM技能关键字，以及一种新的表示学习方法，该方法联合建模图形和文本数据。它将形式化和解决任务和工人之间的双边匹配的优化问题。它还将开发一种有效的方法来识别需要更新技能的OLM工人，并开发一种可解释的启发式方法来为选定的工人提供技能建议。该项目的成果将以同行评审的出版物、会议报告、研讨会和讲习班的形式传播。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Online labor markets (OLMs), where the employment of on-demand workers and the delivery of products occur online, have become an increasingly important component of today’s economy. These two-sided markets provide workers and employers with the access to a large pool of geographically dispersed, semi-anonymous virtual transaction partners. Millions of Americans are freelancing at various OML platforms, and they contribute hundreds of billions of dollars to the U.S. economy. To support workers and employers in performing daily business, OLM platforms have developed some functions and tools. Examples include keyword-based search for jobs, task matching, and recommendation for workers. However, many of these tools have critical drawbacks and cause significant problems (such as bias in the exposure of tasks to workers). This project will collect and analyze massive OLM data and develop a suite of novel technical solutions to improve the efficiency of today’s OLM platforms. The developed technical approaches will make research contributions to the fields of data mining, machine learning, and OLM analytics. The study results of this project will advance the knowledge and understanding of the U.S. OLMs. This project is designed to benefit multiple stakeholders and yield potential impacts on society such as providing a novel worker-task matching tool for OLM platforms, updating OLM workers’ skills, and improving OLM employment. This project integrates research with education through course module development, involving graduate and undergraduate students in research, and research showcases for local K-12 students.This project focuses on the following three specific aims (SAs): developing novel technical solutions for discovering useful knowledge, learning feature representations from massive OLM data, studying the issues of over-competition among workers, mitigating unbalanced exposure of tasks in OLMs, developing a novel two-sided matching approach to address the issues, analyzing the critical skill gap between OLM tasks and workers, and developing a novel forward-looking OLM skill recommendation solution to bridge the gap. The project will develop a context-aware deep & wide approach for mining OLM skill keywords from massive text and a new representation-learning approach that jointly models graphical and textual data. It will formalize and solve an optimization problem for two-sided matching between tasks and workers. It will also develop an effective approach for identifying OLM workers who need skill updating and an interpretable heuristic approach for generating skill recommendations to selected workers. The results of this project will be disseminated in the form of peer-reviewed publications, conference presentations, seminars, and workshops.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.