In this final session of the Internet Archive’s digital humanities expo, Library as Laboratory, attendees heard from scholars in a series of short presentations about their research and how they’re using collections and infrastructure from the Internet Archive for their work.
- Forgotten Histories of the Mid-Century Coding Bootcamp, [watch] Kate Miltner (University of Edinburgh)
- Japan As They Saw It, [watch] Tom Gally (University of Tokyo)
- The Bibliography of Life, [watch] Rod Page (University of Glasgow)
- Q&A #1 [watch]
- More Than Words: Fed Chairs’ Communication During Congressional Testimonies, [watch] Michelle Alexopoulos (University of Toronto)
- WARC Collection Summarization, [watch] Sawood Alam (Internet Archive)
- Automatic scanning with an Internet Archive TT scanner, [watch] Art Rhyno (University of Windsor)
- Q&A #2 [watch]
- Automated Hashtag Hierarchy Generation Using Community Detection and the Shannon Diversity Index, [watch] Spencer Torene (Thomson Reuters Special Services, LLC)
- My Internet Archive Enabled Journey As A Digital Humanities Citizen Scientist, [watch] Jim Salmons
- Web and cities: (early internet) geographies through the lenses of the Internet Archive, [watch] Emmanouil Tranos (University of Bristol)
- Forgotten Novels of the 19th Century, [watch] Tom Gally (University of Tokyo)
- Q&A #3 [watch]
Links shared during the session are available in the series Resource Guide.
WARC Collection Summarization
Sawood Alam (Internet Archive)
Items in the Internet Archive’s Petabox collections of various media types like image, video, audio, book, etc. have rich metadata, representative thumbnails, and interactive hero elements. However, web collections, primarily containing WARC files and their corresponding CDX files, often look opaque. We created an open-source CLI tool called “CDX Summary”  to process sorted CDX files and generate reports. These summary reports give insights on various dimensions of CDX records/captures, such as, total number of mementos, number of unique original resources, distribution of various media types and their HTTP status codes, path and query segment counts, temporal spread, and capture frequencies of top TLDs, hosts, and URIs. We also implemented a uniform sampling algorithm to select a given number of random memento URIs (i.e., URI-Ms) with 200 OK HTML responses that can be utilized for quality assurance purposes or as a representative sample for the collection of WARC files. Our tool can generate both comprehensive and brief reports in JSON format as well as human readable textual representation. We ran our tool on a selected set of public web collections in Petabox, stored resulting JSON files in their corresponding collections, and made them accessible publicly (with the hope that they might be useful for researchers). Furthermore, we implemented a custom Web Component that can load CDX Summary report JSON files and render them in interactive HTML representations. Finally, we integrated this Web Component into the collection/item views of the main site of the Internet Archive, so that patrons can access rich and interactive information when they visit a web collection/item in Petabox. We also found our tool useful for crawl operators as it helped us identify numerous issues in some of our crawls that would have otherwise gone unnoticed.
More Than Words: Fed Chairs’ Communication During Congressional Testimonies
Michelle Alexopoulos (University of Toronto)
Economic policies enacted by the government and its agencies have large impacts on the welfare of businesses and individuals—especially those related to fiscal and monetary policy. Communicating the details of the policies to the public is an important and complex undertaking. Policymakers tasked with the communication not only need to present complicated information in simple and relatable terms, but they also need to be credible and convincing—all the while being at the center of the media’s spotlight. In this briefing, I will discuss recent research on the applications of AI to monetary policy communications, and lessons learned to date. In particular, I will report on my recent ongoing project with researchers at the Bank of Canada that analyzes the effects of emotional cues by the Chairs of the U.S. Federal Reserve on financial markets during congressional testimonies.
While most previous work has mainly focused on the effects of a central bank’s highly scripted messages about its rate decisions delivered by its leader, we use resources from the Internet Archive, CSPAN and copies of testimony transcripts and apply a variety of tools and techniques to study the both the messages and the messengers’ delivery of them. I will review how we apply recent advances in machine learning and big data to construct measures of Federal Reserve Chair’s emotions, expressed via his or her words, voice, and face, as well as discuss challenges encountered and our findings to date. In all, our initial results highlight the salience of the Fed Chair’s emotional cues for shaping market responses to Fed communications. Understanding the effects of non-verbal communication and responses to verbal cues may help policy makers improve upon their communication strategies going forward.
Digging into the (Internet) Archive: Examining the NSFW Model Responsible for the 2018 Tumblr Purge
Renata Barreto (University of California Berkeley)
In December 2018, Tumblr took down massive amounts of LGBTQ content from its platform. Motivated in part by increasing pressures from financial institutions and a newly passed law — SESTA / FOSTA, which made companies liable for sex trafficking online — Tumblr implemented a strict “not safe for work” or NSFW model, whose false positives included images of fully clothed women, handmade and digital art, and other innocuous objects, such as vases. The Archive Team, in conjunction with the Internet Archive, jumped into high gear and began to scrape self-tagged NSFW blogs in the 2 weeks between Tumblr’s announcement of its new policy and its algorithmic operationalization. At the time, Tumblr was considered a safe haven for the LGBTQ community and in 2013 Yahoo! bought Tumblr for 1.1 billion. In the aftermath of the so-called “Tumblr purge,” Tumblr lost its main user base and, as of 2019, was valued at 3 million. This paper digs into a slice of the 90 TB of data saved by the Archive Team. This is a unique opportunity to peek under the hood of Yahoo’s open_nsfw model, which experts believe was used in the Tumblr purge, and examine the distribution of false positives on the Archive Team dataset. Specifically, we run the open_nsfw model on our dataset and use the t-SNE algorithm to project the similarities across images on 3D space.
Japan As They Saw It (video)
Tom Gally (University of Tokyo)
“Japan As They Saw It” is a collection of descriptions of Japan by American and British visitors in the 1850s and later. Japan had been closed to outsiders for more than two centuries, and there was much curiosity in the West about this newly accessible country. The excerpts are grouped by category—Land, People, Culture, etc.—and each excerpt is linked to the book where it first appeared at the Internet Archive. “Japan As They Saw It” can be read online, or it can be downloaded as a free ebook.
Forgotten Novels of the 19th Century (video)
Tom Gally (University of Tokyo)
Novels were the binge-watched television, the hit podcasts of the 19th century—immersive, addictive, commercial—and they were produced and consumed in huge numbers. But many novels of that era have slipped through the cracks of literary memory. “Forgotten Novels of the 19th Century” is a list of fifty of those neglected novels, all waiting to be discovered and read for free at the Internet Archive.
Forgotten Histories of the Mid-Century Coding Bootcamp
Kate Miltner (University of Edinburgh)
Over the past 10 years, Americans have been exhorted to “learn to code” in order to solve a series of entrenched social issues: the tech “skills gap”, the looming threat of AI and automation, social mobility, and the underrepresentation of women and people of color in the tech industry. In response to this widespread discourse, an entire industry of short-term intensive training courses– otherwise known as coding bootcamps– have sprung up across the US, bringing in hundreds of millions of dollars in revenue a year and training tens of thousands of people. Coding bootcamps have been framed as a novel kind of institution that is equipped to solve contemporary problems. However, materials from the Internet Archive show us that, in fact, a similar discourse about computer programming and similar organizations called EDP schools existed over 70 years ago. This talk will showcase materials from the Ted Nelson Archive and the Computerworld archive to showcase how lessons from the past can inform the present.
The Bibliography of Life
Roderic Page (University of Glasgow)
The “bibliography of life” is the aspiration of making all the taxonomic literature available so that for every species on the planet we can find its original description, as well as track how our knowledge of those species has changed over time. By combining content from the Internet Archive and the Wayback Machine with information in Wikidata we can make 100’s of thousands of taxonomic publications discoverable, and many of these can also be freely read via the Internet Archive. This presentation will outline this project, how it relates to efforts such as the Biodiversity Heritage Library, and highlight some tools such as Wikicite Search and ALEC to help export this content.
Automatic scanning with an Internet Archive TT scanner (video)
Art Rhyno (University of Windsor)
The University of Windsor has set up a mechanism for automatic scanning with an Internet Archive TT scanner, used for the library’s Major Papers collection.
Automated Hashtag Hierarchy Generation Using Community Detection and the Shannon Diversity Index
Spencer Torene (Thomson Reuters Special Services, LLC)
Developing semantic hierarchies from user-created hashtags in social media can provide useful organizational structure to large volumes of data. However, construction of these hierarchies is difficult using established ontologies (e.g. WordNet) due to the differences in the semantic and pragmatic use of words vs. hashtags in social media. While alternative construction methods based on hashtag frequency are relatively straightforward, these methods can be susceptible to the dynamic nature of social media, such as hashtags associated with surges in popularity. We drew inspiration from the ecologically-based Shannon Diversity Index (SDI) to create a more representative and resilient method of semantic hierarchy construction that relies upon graph-based community detection and a novel, entropy-based ensemble diversity index (EDI) score. The EDI quantifies the contextual diversity of each hashtag, resulting in thousands of semantically-related groups of hashtags organized along a general-to-specific spectrum. Through an application of EDI to social media data (Twitter) and a comparison of our results to prior approaches, we demonstrate our method’s ability to create semantically consistent hierarchies that can be flexibly applied and adapted to a range of use cases.
Web and cities: (early internet) geographies through the lenses of the Internet Archive
Emmanouil Tranos (University of Bristol)
While geographers first turned their focus on the internet 25 years ago, the wealth of data that the Internet Archive preserves and offers remains at large unexplored, especially for large projects in terms of scope and geographical scale. However, there is hardly any other data source that depicts the evolution of our interaction with the digital and, importantly, the spatial footprint of this interaction better than the Internet Archive. Therefore, the last few years we have been using extensively data from the Internet Archive in order to understand the geography and the evolution of the creation of online content and their interrelation with cities and spatial structure. Specifically, we have worked with The British Library and utilised the JISC UK Web Domain Dataset (1996-2013)1 for a number of projects in order to (i) explore whether the availability of online content of local interest can attract individuals online, (ii) assess how the early engagement with web tools can affect future productivity, (iii) map the evolution of economic clusters, and (iv) predict interregional trade flows. The Internet Archive helps us not only to map the evolution and the geography of the engagement with the internet especially at its early stages and, therefore, draw important lessons regarding new future technologies, but also to understand economic activities that take place within and between cities.