Extraction of Citation Data from Websites based on Visual Cues

Every scientific article, be it an essay or a paper in a journal, provides citations and a bibliography to support its arguments. Publishers and universities require their academics to format their bibliography using predefined styles like APA, MLA or Harvard, to name just a few. Obviously it is a tedious job to do that by hand, which is why there is a multitude of software to format the references following style guides, provided the required information is present in a structured format. Some services (citation generators), like CiteThisForMe1, EasyBib2 or RefME3 even try to provide this information automatically, so the user only needs to enter an identifier like an ISBN, DOI or URL to add a specific work to the bibliography. This thesis focuses on the case, where a web resource needs to be referenced. One would have to open the website and find the relevant information, such as the title, author, date of publication and publisher, needed to reference this specific resource. In the scope of this work, this information will be called citation data. The aforementioned citation generators automate this process, so that one only needs to enter a URL. The software tries to extract data required to cite this article. For example, to reference a news article on BBC online, the citation generator 1) would open the URL, 2) extract the title, author, date of publication and publisher, and 3) finally return a properly formatted reference according to the selected style. In my work for RefME, a start-up creating a citation management platform, I created and enhanced the underlying scraper service, which, given a URL, fetches the HTML code of that website and, based on hard-coded identifiers, extracts citation data. This method is very sensitive to the way a website’s code is written, which inspired a different approach. Based on the experience gathered, it seems that using visual cues might improve the precision of the citation data extraction. The primary advantage is, that those are independent