Networks of Net Literature - Modelling, Extracting and Visualizing Link-Based Networks in the DLA corpus of net literature

Net literature on the WWW is characterized by a special relationship between literary text and technical medium. In addition to the importance of graphic and typographic design, this includes in particular the hypertext structure of the texts. The distribution of a literary text across several interlinked web pages often leads to a non-linear structure, which usually corresponds to non-linear narrative structures. We consider a non-linear text structure to exist as soon as a page contains more than one reference to subsequent pages. A linear passage through the entire text is then no longer possible. For narrative texts, this also implies a non-linear narrative progression. Non-linear text structures can allow predominantly linear narrative progressions with alternative strands, variable endings, and cyclical elements, or multiple narrative progressions through complex linking.
For the identification of non-linear patterns, we extract link and element structures from the corpus data and visualize the link networks. We then identify patterns via the visualization and network metrics. A necessary prerequisite for corpus-oriented pattern recognition is the comparability and reproducibility of the individual analyses. Live online literary works can change over time. We work with archived versions of the works, which are kept available via a repository. The pages are archived in the WARC format, a standard format for web archiving.
We have modelled and described all types of links we found in the corpus. This model is mainly based on our historical corpus and serves to improve extraction and analysis, but will be tested and complemented with additional material in the future. Additionally, we assessed several existing approaches for the extraction of links. For the actual extraction of the link network data from the WARC files, we use our own software module WARC2graph which is based on our link network model and makes use of multiple extraction approaches. WARC2graph extracts link networks from WARC files and returns these link networks. Users can choose which method to use to extract the network data. The program has a basic module for generic visualization of the data, but this is just for a first impression, since visualization depends on the research question and the nature of the analyzed material.
WARC2graph will be made available as a Python module and as a web service run by the Science Data Center for Literature, a joint project of the German Literature Archive (DLA) and several departments at the University of Stuttgart.
Our approach is focused on net literature and literary blogs. However, it should work generically on standard WARC files. We hope to make the module reusable both at the level of research-oriented work with WARC data and at the level of archiving and provision of WARC data and data analytics services.