Collaborative Research: III: Small: Robust Learning and Inference Protocols for Mitigating Information Pollution

点击翻译按钮获取中文摘要

Social computing platforms, in which every user is a potential information source, have revolutionized the way content is generated and disseminated in multiple fields, such as news, healthcare and online education, to name a few. In these open settings, users often access information without awareness of its source and its level of expertise making them more susceptible to manipulation and exposure to biased or even deceptive content. This project is designed to help users navigate the information space, by defining and operationalizing the concept of Information Pollution, the contamination of information supply with irrelevant, redundant, unsolicited, incorrect, and otherwise low-value information, and suggest principled methods for combating its adverse effects by augmenting it with the relevant context. The framework will identify the spectrum of perspectives that could exist around topics of public interest, identify relevant expertise, and thus improve public access to diverse and trustworthy informationThe goal of this project is to address some of the key research questions in support of mitigating information pollution. The project’s approach decomposes the problem into its core components – from key natural language processing problems that arise when attempting to identify and present the multiple perspectives a claim might have, along with its supporting evidence, to understanding information sources, the claims they make and evidence they provide, to an algorithmic inference framework for trustworthiness. The investigators will define novel learning and inference tasks that would provide important building blocks for addressing the information pollution problem. These include a holistic framework for assessing trustworthiness of information by consolidating information about sources from multiple documents and their interactions with other sources. In addition, the project will define novel language understanding tasks which provide new abstractions supporting the characterization of the similarities and differences between claims, the intent behind them, perspectives they express, and their implications. The research goals will be complemented by a comprehensive evaluation plan, consisting of both intrinsic evaluation of each capability, as well as extrinsic evaluation measuring the impact the full system will have on information consumed by users.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.