Dynamic Classification in Web Archiving Collections

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1459–1468 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1459 Dynamic Classification in Web Archiving Collections Krutarth Patel, Cornelia Caragea, Mark E. Phillips Kansas State University, University of Illinois at Chicago, University of North Texas kipatel@ksu.edu, cornelia@uic.edu, Mark.Phillips@unt.edu Abstract The Web archived data usually contains high-quality documents that are very useful for creating specialized collections of documents. To create such collections, there is a substantial need for automatic approaches that can distinguish the documents of interest for a collection out of the large collections (of millions in size) from Web Archiving institutions. However, the patterns of the documents of interest can differ substantially from one document to another, which makes the automatic classification task very challenging. In this paper, we explore dynamic fusion models to find, on the fly, the model or combination of models that performs best on a variety of document types. Our experimental results show that the approach that fuses different models outperforms individual models and other ensemble methods on three datasets. Keywords: Web Archiving, Machine Learning, Deep Learning, Dynamic Classification 1. Introduction A growing number of research libraries, museums, and archives around the world are embracing Web Archiving as a mechanism to collect born-digital material made available via the Web. Between the membership of the International Internet Preservation Consortium, which has 55 member institutions (Consortium, 2017), and the Internet Archive’s Archive-It Web Archiving platform with its 529 collecting organizations (Archive-It, 2017), there are well over 584 institutions currently engaged in building collections with Web Archiving tools. The amount of data that these Web Archiving initiatives generate is typically at levels that dwarf traditional digital library collections. As an example, in a recent impromptu analysis, Jefferson Bailey of the Internet Archive noted that there were 1.6 Billion PDF files in the Global Wayback Machine. If just 1% of these PDFs are of interest for collection organizations, that would result in a collection of 15+ million volumes in HathiTrust. While the number of Web Archiving institutions increases, the technologies needed to provide access to these large collections have not improved significantly over the years. At this time, the standard way of accessing web archives is with known URL lookup using tools like the OpenWayback1 or pywb2. The use of full-text search has increased in many web archives around the world, but often provides an experience that is challenging for users because of the vast amount of content and the limitations of strictly text- based searches for these large heterogeneous collections of content. Another avenue of access to web archived data that is of interest to Web Archiving institutions is the ability to extract high-quality, content-rich publications from the web archives in order to add them to their existing collections. In this paper, we explore machine learning and deep learning models to classify documents from Web Archiving collec- tions into being in scope for a given collection (or collection policy) or …

Content truncated to 500 words