A timeline of training experiences and "ah hah" moments leading to my digital footprint: Late 1980s: Two key Boston University professors shaped my double major in English and Political Science. The English professor, Susan Mizruchi, taught me to read works very closely with a set of colored pens nearby. Underlining, writing in the margins, and creating memos were all fundamental pen and paper techniques that later became parts of my digital methods. Professor Mizruchi also pointed me to primary historical document research. She taught me to write with care. The Political Science professor was a 1960s political activist, WWII pilot, and historian named Howard Zinn. There are so many stories about Zinn, it is hard to know where to begin. His classes required students to join a social movement of their choice. We participated in our free time and journaled about it. I helped organize a train to the April 25, 1987 "March on Washington" to protest Apartheid and the US role in wars in Central America. He emphasized reading and writing to newspapers as forms of homework. Together, the two of them advised my honors thesis on political extremism and the novel in which primary historical sources, inductive qualitative methods, and as ever in research, chance discoveries, led me to write about four pairs of archetypal characters across two books ("The Secret Agent" Joseph Conrad and "Boston" Upton Sinclair). When I think back to everything that happened later, it is impossible not to see this year of my undergraduate life as fundamental to my current digital methods. It was in this phase I first learned that when you find something interesting and unexpected, there is value in looking for all the other examples. We never called it "coding" or even a methodology, but it was an inductive qualitative method leading to more structured color coded labeling of book passages over time. Early 1990s: I spent two and a half years in the University of Oregon English PhD program. As a funded PhD student, I was part of a small army that taught Writing 101 for the department. The core methods taught by my three favorite professors were literary theory, feminist theory, postmodernism, and semiotics. By the time I realized I could never pass the required Old English class, I dropped out and became a Tilth certified organic farmer while squatting on farmland previously owned by the Marcos family. Stay with me here. In organic farming, you are constantly learning from trial and error how and when to plow, dig, build, water, weed, plant seeds, harvest, and a variety of essential tasks to get fresh produce to the Eugene Farmers Market. It reinforced my view of learning by doing. I had no farming mentor, just a green thumb that needed an opportunity to express myself. Mid-Late 1990s: I returned to the University of Oregon to complete a PhD in Political Science. The field was heavily quantitative and some parts were turning to game theory, rational choice, and experimental design. The required methods sequence was stats, econometrics, and applied econometrics. I survived (barely) a bruising journey through these classes where I lost confidence in my ability to use the "state of the art" methods required to be in the mainstream of my discipline. My graduate advisor, Ron Mitchell, helped purge the remnants of postmodernism from my writing and turned my head toward the scientific method parts of political science. In this moment, the first chapter of King, Keohane and Verba's "Designing Social Inquiry" was the key to unlocking the login of inference that cuts across quantitative and qualitative methods. My thesis chair was a "Politics and History" section member, Gerald Berk. One of the key committee members was Jeffrey Ostler, a journalism history professor. Both of these professors were united around the importance of immersing yourself in primary documents. My primary data source (newspapers from 1907-1916) came from inter-library loan wrapped carefully in brown butcher paper. For example, I had the first four years of the newspaper of the US Chamber of Commerce. There was a remarkable publication called "The Banker-Farmer." The methods? Carefully unpack all of the papers and scan every page for stories about the problem of farm credit. What happens when you spend week after week flipping through old newspapers is you find many things you were not looking for that help you understand and better explain, in my case, how the government got into lending money to farmers in 1916. For example, the "cost of living" in cities was a pervasive topic driving the need for better farming. More important to me, coming off three years of farming, was the surprise discovery of a "soil fertility" movement that was a response to poor farming by tenants who had no access to reliable or fair credit. These concepts became "codes" in my manual labeling of 600+ articles photocopied from the old papers. When Congress finally passed the law in 1916, it was the culmination of years of agenda-setting and framing in the print press. It also borrowed a banking model from Europe (cooperative farm credit banks) that established the first 30-year mortgages in the U.S. If you have ever had a 30-year mortgage, you should be grateful to the editor of The Banker-Farmer and other urban agrarians. 1999: As a first year visiting assistant professor of political science at Providence College, I submitted more than 140 paper job applications. This involved a team of student assistants and a forest of paper. It was a tough market for a political historian. I needed a new topic and a new methodology. I saw a news story. The US Department of Agriculture was promulgating a new national standard for the concept of "organic" agriculture and it caused a major backlash. Oregon Tilth was the nation's gold standard at the time. Sewage sludge, irradiation, and genetic engineering were proposed as acceptable organic farming practices. The agency received almost 250,000 public comments. Some were postcards and printed mail. However, 20,000 came in emails and the agency personnel met with me and gave me a copy of the electronic comments on a 100MB Zip Disk. Sensing unemployment was a real possibility, my academic father helped me search the "Community of Science" for grants. We found the newly established "Digital Government" program at the NSF and 30 days later I had a seed grant of $30,000 to assess whether agencies could use NUD*IST (now NVivo) to review large scale public comment submissions. The short answer was no. Having 20,000 documents required something else. The program manager wanted to know what would happen if there were 1,000,000 comments with a 90-day agency review period to find and address the important comments in the final rule. 2000-2004: I was a founding assistant professor of Environmental Science and Policy at Drake University. This was an interesting moment because the interdisciplinary program was not as common as they are now. Despite being a liberal arts college, I was strongly encouraged to pursue the funded research, which included a large NSF grant to see if service-learning could bridge the digital divide. It could not. 2000-2008: Through the Digital Government program I was encouraged by the program manager to write proposals involving computer scientists. Working with Jamie Callan (Information Retrieval) and Eduard Hovy (Natural Language Processing) we launched an eRulemaking Research Group that also included a sociologist, a political theorist, a statistician, and me, the political historian. Agencies started giving us bigger datasets (ex., 500,000 electronic comments on the mercury rule from EPA) which we posted in a public "testbed" at CMU because these were novel the same way the Enron emails were. Another Digital Government-funded researcher, David Lazer, popularized the term "Computational Social Science" in the journal Science in 2009. The Digital Government program explicitly called for deeper collaboration between the disciplines and involvement by federal agency personnel. Our group did a series of transcribed interviews and focus groups in 2003 and 2004 asking a variety of stakeholders, academics, law professors, business leaders, citizen activists, and agency officials what the biggest problems might be in the era of eRulemaking. The most common fear was information overload. 2004-2007: Over this period my academic work included a lab at Pitt: the Qualitative Data Analysis Program (QDAP). This was a fee for service space where teams of annotators labeled data from a wide variety of disciplines using Atlas.ti. The computer science professors and students in the eRulemaking Research Group needed to measure inter-rater reliability. Atlas.ti did not support this at the time and even opposed it on methodological grounds. This is where everything started to change. One of Eduard's students built a "kappa tool" which was a web-based tool for calculating agreement when multiple annotators label the same data using Atlas. It was a very basic utility on the web and people used it. She graduated and there was nobody to take over the tool. I was very fortunate to have a pot of indirect cost recovery funds. In August 2007, I started working with a CMU engineer employed by my collaborator in his LTI lab. Next August will be the 20th anniversary of our work together. Mark Hoy developed the first 25,000 lines of software code as a free, open source, web-based Coding Analysis Toolkit over the first year. It was originally just a tool for ingested coded Atlas datasets. We started adding features for crowdsource annotation on the CAT platform. Perhaps most important, we developed an adjudication module for resolving annotator disagreement. This is by far the most important and least used tool in our journey. If you book a meeting and want to discuss adjudication, it is my favorite topic in the annotation universe. 2009-2012: CAT evolved into the commercial Public Comment Analysis Toolkit (PCAT) involving invention disclosures to Pitt and UMass Amherst, where I worked 2008-2012 and also a QDAP lab, this one was hosted by the UMass Computer Science department. In 2010, PCAT became DiscoverText. The first commercial customers were JetBlue and Google. They had large surveys and our tools have proven effective for surveys with thousands or tens of thousands of open ended answers. The first major data and methods customer in 2012 was a CDC-funded smoking study where we first established access to the Gnip full fire hose of Twitter data and novel solutions for the problem of word sense disambiguation at scale with social media data. The five pillars of text analytics we focus on emerged in this period: search, filter, cluster, code, and machine-classify text. Each tool set facilitates unique ways to do inductive research, which remains my overarching approach despite the "power tools" for text replacing the colored pens from the 1980s and 1990s. 2013-2014: A multinational market research firm bought the software and Mark & I went to work for them. Looking back, I think they thought ML in 2012 was AI in 2026. It was not. These were tools to shorten the time it takes to build a reusable custom machine classifier. Employees in the firm were not consulted before we were brought in and it became clear there was a mismatch. They wanted magic buttons and not a scientific research platform requiring learning new methods and doing actual work. Fourteen months later I had ownership of the software code back and the firm kept my engineer, sort of. 2014-2018: We operated Sifter, which was a self-serve, pay-as-you-go, platform for accessing the complete history of Twitter. We were official partners of the company Gnip that Twitter bought for 140M USD because they did archival search better than Twitter, which struggled to keep itself running at times. With Sifter, you set the date range, plugged in a query, got the estimate, adjusted and repeated until ready, swiped a credit card, and data flowed directly to your DiscoverText account. Users generated about 16,000 estimates using Sifter. We also sold a perpetual license in 2018 to the U.S. Fish and Wildlife Service who we had been supporting continuously since the 2007 polar bear rulemaking. They had funded the prototype that became PCAT then DiscoverText. 2019-2025: Various disruptions to the Twitter data ecosystem happened. Many people have their own stories. The data itself disrupted my life. Finding QAnon in August 2019 meddling in the Canadian election led me down a variety of data-driven rabbit holes. Mark built a tool for bot detection. I started learning more about the weaponization of social media. The intel community had me give a solo brief to the Strategic Multilayer Assessment on Feb 12, 2020. The talk, "Hunting Bots and Trolls in a Game Environment," had me pretty far afield from journalism history but still using my own peculiar form of inductive strategies to warn about the "digital soldiers" who went on to plan and execute J6. The 2021 insurrection was planned in the open using public facing Tweets and strategies like the "Twitter Train" to tag up to 49 people in threads that served as notification engines. Watching the events of that day and the subsequent role of digital soldiers reviving and re-electing President Trump caused my family to leave the United States fearing the campaign of retribution. 2026: Texifter, LLC shuts down in the US and Texifter, Inc. launches a re-startup in Vancouver, BC. We moved almost 1 million lines of software code to Canada. Over the last six months we have been in good negotiations with X and have started to restore approved academic and market research access to the X API via a new product: XSifter. Currently you can query for an estimate that goes back 30-days. Academics can license X data for research via XSifter for analysis in DiscoverText. This was approved by X senior managers for reasons I have emphasized in posts to the list over the years. We built compliance into the platform in 2012. We use the Twitter/X official display. We do not display deleted, suspended, or protected Tweets. The data is authentic, bearing the hallmarks of the live display, and does not come as flat text in CSV or JSON format. We are also introducing AI-enhanced query building and microsampling that improves the ability of the researcher to refine and test queries before spending money on data. If and when an institution or set of organizations becomes an Enterprise customer of XSifter, access to the full history of public Twitter will be restored for everyone in academia willing to pay to license data. The current price point we offer to enter the X data license market is $100 USD. You can no longer say $42,000/month is the entry level for API access. So, to sum up, this is my on-the-job journey from journalism historian to methodologist to software founder. The unifying thread is finding unexpected concepts inductively and never making deductive expectations about the data determine my research options. In the process of following the data, I always get ideas for new software tools and methods. Generally speaking, if they pass the sniff test, Mark builds them. One million lines of code later, he is doing more innovative and interesting software engineering than ever. If you have never tried our tools, I am curious why. If you have never tried our methods, you might be missing out on a small piece of the history of computational social science. https://tinyurl.com/DiscoverTextGoogleScholar On Mon, Nov 2, 2026 at 3:06 AM Salomé Hoarau via Air-L <air-l@lists.aoir.org> wrote:
Since you all investigate the internet from various methodological angles, I am very curious about your own professional experiences regarding your digital footprints:
- Did you ever receive formal institutional training or dedicated seminars during your graduate studies on how to manage your own online presence/digital footprint as a researcher? - Or is this something you mostly learned implicitly, "on the job," or through peer-to-peer advice?