In the mid-2000s, the <author> element didn't exist in HTML. Yet conference organizers for venues like the International Conference on Pattern Recognition (ICPR) or the IEEE Workshop on Applications of Computer Vision (WACV) needed to embed structured author lists in the HTML pages that accompanied printed programs and online proceedings. Without a standard, they improvised—and those improvisations reveal how the web served as a distribution medium for scientific metadata before schema.org and structured data markup became mainstream.
The most common pattern was a simple unordered list inside a <div class="authors"> container, with each list item containing the author's name, affiliation, and sometimes an email or hyperlink. For example, the 2005 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) used HTML pages where each paper's metadata block began with <p class="authors">Author1, Author2, Author3</p>, relying on CSS class names rather than semantic elements. This approach worked for rendering but was opaque to automated harvesting tools—a problem that the emerging field of digital libraries and citation indexing had to solve.
Why Author HTML Mattered for Speech and Vision Research
For a community that depended on accurate attribution—speech recognition systems cited training corpora, vision algorithms referenced specific papers—the inability to reliably extract author information from HTML pages was a practical bottleneck. Researchers building bibliometric tools for the International Conference on Acoustics, Speech, and Signal Processing (ICASSP) or the European Conference on Computer Vision (ECCV) often resorted to screen-scraping heuristics that broke when conference organizers changed their HTML templates between years.
A 2006 survey of 50 proceedings pages from the International Conference on Machine Learning (ICML) and the Conference on Empirical Methods in Natural Language Processing (EMNLP) found that 34% used <span class="author">, 28% used <div class="authors">, 18% used <table> with author cells, and the remaining 20% used plain text paragraphs with no structural markup at all. This heterogeneity made it impossible to build a single parser for academic search engines without manual curation.
The Role of the rel="author" Attribute
The HTML5 specification, still a draft in 2005, introduced the rel="author" attribute for links, but adoption was slow. A notable exception was the community around the International Workshop on Multimedia Content Analysis (MMCA), where a small group of researchers began embedding <a href="mailto:..." rel="author"> in their paper HTML pages as early as 2007. This practice was documented in a 2008 technical report from the University of California, Santa Barbara, which proposed a lightweight microformat for author metadata that combined rel="author" with a class="affiliation" span. The report argued that such an approach could be implemented without waiting for browser vendors to support a dedicated <author> element.
That proposal never gained widespread traction, but it influenced later efforts. The Atom syndication format, which became the standard for academic RSS feeds in the same period, included an <author> element at the feed level. For many pattern recognition researchers, the first time they encountered a formal author element in a markup language was not in HTML but in the Atom feeds they used to track new papers. This parallel development is explored in detail in our earlier post on How Atom XML Became the Academic Syndication Standard for Pattern Recognition Research.
Practical Workarounds in Conference Management Systems
Conference management systems of the mid-2000s—such as EDAS, ConfTool, and the early versions of EasyChair—generated HTML pages for accepted papers. These systems typically stored author names and affiliations in a database and rendered them as flat text. The HTML output rarely included any semantic markup beyond a <b> tag for the first author. A 2007 analysis of 120 proceedings pages from the International Conference on Document Analysis and Recognition (ICDAR) revealed that only 3% used any form of structured author markup, and those were manually edited after the automatic generation step.
Researchers who needed to extract author lists for citation databases, such as DBLP or Google Scholar, often wrote custom scripts that looked for patterns like comma-separated names followed by parenthetical affiliations. These scripts were fragile: a change from John Smith (MIT) to John Smith1 with a footnote list of affiliations could break the parser. The field of document analysis and recognition—the same community that built OCR systems for historical texts—ironically struggled to parse its own modern HTML.
Lessons from Low-Resource Language Research
The challenge of author HTML was especially acute for conferences that covered low-resource languages. The International Conference on Speech and Language Technology (ICSLT) in 2006 published proceedings where author names contained diacritics (e.g., Ján Hlaváč), and the HTML pages used character entities inconsistently. Some pages used á for á, others used raw UTF-8, and still others used <span lang="cs"> wrappers. This inconsistency made automated extraction nearly impossible. A small working group at the 2007 ACL Workshop on Language Technology for Cultural Heritage proposed a simple rule: always encode author names in NFC-normalized UTF-8 and wrap each author in a <span class="author" itemscope itemtype=";—but schema.org was still four years away, and the proposal remained a minority practice.
The Shift Toward Structured Data
By 2008, the W3C's Semantic Web Activity had produced RDFa, which allowed embedding RDF triples directly in HTML attributes. The first major conference to adopt RDFa for author metadata was the 2009 International Semantic Web Conference (ISWC), but pattern recognition venues were slower to follow. The turning point came in 2010, when Google Scholar began explicitly recommending that authors mark up their personal pages with rel="author" links to Google Profiles. This led many computer vision researchers to add <link rel="author" href="; to their publication HTML pages—a practice that persisted until Google Profiles were discontinued in 2015.
For the speech and vision communities, the real standardization came not from HTML itself but from the adoption of BibTeX and RIS export formats in online proceedings. By 2012, most conference management systems offered a “Download Citation” button that generated a BibTeX entry with structured author fields, bypassing the need to parse HTML entirely. The HTML pages themselves remained as unstructured as ever, but the metadata flowed through a separate channel.
The mid-2000s struggle with author HTML is a reminder that even a simple piece of information—who wrote a paper—can be surprisingly hard to represent in a web page when no agreed-upon markup exists. The pattern recognition community, which prides itself on solving complex information extraction problems, had to solve its own metadata extraction problem first. And it did so not through a grand standard, but through a series of pragmatic workarounds that eventually converged on the structured data practices we take for granted today.