FMitF: Track II: Programming by Demonstration for the Browser with Applications in Data Science

点击翻译按钮获取中文摘要

Future work will revolve around consuming data, deriving insights, and making data-informed decisions. While data analysis and machine learning can be significantly automated, an irreducibly human part of the data-driven future is data collection and curation. Currently, data collection is done either by experienced engineers or through manual data collection. Programming by Demonstration (PBD) has recently allowed non-experts to author automatic web-scraping programs, which collect data efficiently and reliably. This project will extend a previously developed system for browser-based PBD and evaluate it in a large-scale deployment. The scale will be large both in terms of users and the size of data that will be collected. The system development part of the project will focus on two parts: (1) Implementing and deploying algorithms for large-scale deployment of the Helena browser-based PBD system. This work will include parallelization of synthesized web-scraper programs, making programs robust to failures of web sites and the network, and adapting the parallel execution to optimize for either the data collection rate or the cost of datacenter time. (2) Developing a next-generation user interface that will include support for debugging the synthesized programs. The target audience of these tools will be teams of data-driven social scientists with whom the PI team has had long-term collaborations. These teams have been using the Helena PBD system for collecting datasets that improve strategies of non-profits and influence government policies. The evaluation will investigate both the system performance and the usability of the novel and critical parts of the Helena ecosystem.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.