Journalism is at the crossroads. In the past, we have come to rely on investigative reporting by traditional news organizations to hold governments, corporations, and individuals accountable to society. In recent years, there has been an alarming trend in the increasing amount of misinformation, compared with stagnant resources and talents devoted to investigative reporting. This trend has a profound impact on the well-being of democracy. At the same time, there is also an opportunity. With technological advances and the movement towards transparency, the amount of data available to the public is ever increasing. However, the potential of this "democratization of data" cannot be realized with the widening divide created by the growth of data far outpacing the investment in investigative journalism.

Computing is a key to bridge this divide. Computational journalism aims at developing computational techniques and tools to increase effectiveness and broaden participation for journalism---especially public interest journalism---to help preserve its watchdog tradition. In this project, we consider how we, as computer science researchers, can contribute to journalism and help promote computational journalism as an emerging discipline.

At Duke, UTexas at Arlington, Stanford, and Google, our interdisciplinary team of computer scientists, journalists, and public policy researchers is currently focusing on computational fact-checking, to help guard against "lies, damned lies, and statistics"---claims that are factually incorrect, or correct but still misleading. We seek to quantify various measures of "goodness" of claims over data, and develop techniques for computing these measures and rebuking misinformation. We are also working on lead-finding, which helps uncover patterns from data that can lead to interesting, robust claims or news stories.

This project is a collaboration with the IDIR Lab at UTexas at Arlington (project website) and the DeWitt Wallace Center for Media and Democracy at Duke University (project website).

People

Computer scientists:

• At Duke: Pankaj Agarwal, Jun Yang
• At UTexas Arlington: Chengkai Li
• At Google: Cong Yu

• iCheck analyzes the voting records of the U.S. Congress and lets you compare how legislators vote with party majorities and the President, and more importantly, explore how the comparison stacks up under different contexts---over time, among groups of peers, and for "key votes" identified by lobbying/political organizations.
• See coverage in Duke Today and Chronicle, or check out our Science Cafe presentation at the North Carolina Museum of Natural Sciences.
• An earlier version of iCheck was demonstrated at SIGMOD 2014. Besides voting records, it also sifted through sports data as well as data on academic publications (DBLP) to make interesting claims and check them.
• ClaimBuster is a machine learning tool that helps find political claims from text sources to fact-­check. It has been watching the 2016 elections in the U.S. as well as the Australian Parliament.
• FactWatcher monitors sports and weather data and automatically finds interesting or newsworthy factoids. It won the Excellent Demonstration Award at VLDB 2014.

Publications

For a general introduction to computational journalism, see the article below by Cohen, Hamilton, and Turner in Communications of the ACM, 2011. Our CIDR 2011 paper (which was the third-place winner of the Best Outrageous Ideas and Vision Track Paper Award) outlined some research challenges from the standpoint of database researchers.

Journalists may find our papers in the Computational+Journalism series over the years helpful.

Database researchers interested in our approach to checking "correct but misleading" claims can read our PVLDB 2014 paper on computational fact-checking and PVLDB 2016 on perturbation analysis.

(If you are having trouble seeing the publications above, please try this link instead.)

Funding Acknowledgments

Our work has been supported by funding from Google, HP, Knight Foundation, and National Science Foundation (on perturbation analysis of data queries).

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the funding organizations.