The ideal ReaderBot:: Machine readers and narrative analytics

Abstract

When Artificial Intelligence is considered in the context of creative activities, it is normally in the role of the creator (e.g. deep learning algorithms applied to poetry or painting). In this paper we instead propose casting the machine in the role of a reader (a ReaderBot), to help authors see how their work was likely to be experienced by an audience. This seems especially useful with interactive narratives, where multiple paths can produce many thousands of variations on that experience. We present an exploratory experiment based on the StoryPlaces locative narrative system, showing how a Simple Heuristic ReaderBot can simulate readings of a located hypertext. We then provide Narrative Analytics of those readings in the form of structural, experiential/dramatic, and locative feedback. We also present three Machine Learning ReaderBots (Linear Regression, Logistic Regression, and a Feed Forward Neural Network), trained on real reading logs, and using distance, prior visits, altitude, proximity to POIs, and text similarity as an input vector to predict next node decisions with precision substantially better than random, and comparable to the Heuristic Reader. We argue that ReaderBots can create an instant audience of thousands that could give authors valuable insights into the potential experiences of their readers.

More information

Identifiers

ORCID for David Millard: orcid.org/0000-0002-7512-2710

Catalogue record

Export record