In this paper we present the Rolling Stylometry Explorer, a web-based tool that displays the results of rolling.classify analyses from Stylo R in the analysed text.

Especially in cases of collaborative or anonymous authorship, it is often difficult to attribute authorship to a text. One approach to authorship attribution using distant reading is stylometric analysis with Stylo, a library in R. (Eder et al. 2016: 107). As a library, Stylo offers a wide range of functions, from simple stylometric cluster analyses to machine learning algorithms. Because of the possibility of using a GUI, Stylo is particularly user-friendly and can also be used by people without in-depth programming knowledge.

For the analysis of collaborative authorships, Stylo offers, among other things, the rolling.classify function. With the Rolling Stylometry Explorer, we present a tool that can be used to visualise the results of rolling.classify directly in the text to be attributed.

For the analysis with rolling.classify, the text to be analysed is divided into possibly overlapping text chunks of equal size (Eder 2016: 457-469). A machine-learning algorithm and a training corpus, for example with text samples of possible candidates for authorship, are then used to classify the individual text sections. Several machine-learning methods are available in Stylo.

The results of the rolling.classify analyses are available both as raw data and as a plot. In both versions, the position of the individual assigned text sections in the overall text is indicated by the number of words only (i.e., a position at 10000-15000 words). One way to alleviate this problem is to manually set chapter or section markers that appear in the plot. This can be done by inserting a keyword into the text to be analysed. Although this enables a rough orientation, it does not solve the core problem: finding the respective text passage in the analysed text which still is only possible by estimating the position on the basis of the scaling or the markers. For a close reading following up on the distant reading, this is cumbersome.

The Rolling Stylometry Explorer solves this problem. It offers a technical solution that ties in as seamlessly as possible with rolling.classify. Thus, the Explorer is located in the intermediate area of stylometry and computer application / data visualisation. It was programmed in Typescript on the basis of Angular. The interface between Stylo and the explorer is programmed in R. Angular Material was used to create the user interface, the core element of the Rolling Stylometry Explorer.

In the Explorer interface, the analysed text is displayed first. The classification made by Stylo is recognisable by a colour highlighting in the text itself. Technically, the assignment is realised by transferring the raw data from rolling.classify into a JSON file, which is then loaded into the Explorer. The possible authors are each assigned a colour. The respective sections are then coloured using the colour of the most probable author for the text segment. It is also possible to display multiple analyses at once, allowing quick comparisons between different settings of Stylo (e.g. different algorithms, MFWs, etc.). This is helpful in validating individual analyses.

The central challenge in developing the explorer was to design and programme a form of presentation that is as close as possible to the plot generated by stylo. The goal here is that the plot and the explorer can complement each other as seamlessly as possible. In order to realise this goal, the explorer uses the same colour allocation as stylo and also displays a Stylo-like mini-map on the right hand side. This makes it easy to navigate through the text. Other features of the Rolling Stylometry Explorer include the possibility of displaying the raw data for each text section via a tooltip. In addition, metadata, for example the names of the possible authors, settings of the individual analyses, etc. can be transferred. With the gradient feature, it is possible to visualise the confidence of the attribution in the text. For each text segment, the colours of e.g. the different authors may also be mixed according to the ratios of the probabilities from the raw data of Stylo. This mixture creates a colour representation of the certainty of the attribution.

The Rolling Stylometry Explorer thus simplifies the interpretation of the raw data for users of Stylo itself. The accessibility of the raw data is also improved through the intuitive representation. Thus, the Explorer also offers the potential to promote interdisciplinary cooperation. Researchers who concentrate mainly on close reading are given the opportunity to include results from corresponding distant reading analyses without having to switch from the text format to a rolling.classify plot. The format of the web component and the ease with which it can be embedded on a website, allows for an interactive presentation of research results that can be shared via a link. Therefore, the Explorer also contributes to science communication in stylometry.

The short paper aims to present the Rolling Stylometry Explorer in both technical and practical terms by discussing the data formats used, as well as illustrating the software by exploring a concrete research topic. The explorer will be introduced and its potential, especially with regard to its use for interdisciplinary projects, will be explained. This is an open source project in which collaboration is always welcome. Possibilities for expansion are also being discussed, such as the development of an Electron app for the Explorer, in order to further optimise user-friendliness and accessibility.