Beyond Canonical Texts - A Computational Analysis of Fanfiction

Aug 27, 2026

User-created fanfiction enables computational study of an ecosystem where literature production, consumption, and social interaction coexist. Using a large-scale dataset from fanfiction.net (55 billion tokens, 2 million users), the work conducts empirical analyses for NLP, computational social science, and digital humanities. Findings show that fanfiction deprioritizes main protagonists versus canonical texts, reveals statistically significant gender differences in character attention, and supports modeling how readers respond to stories.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages6

Reading Time15 min

AccessFree to download

Introduction

Fanfiction data

Data collection and dataset scale

Languages and corpus composition

Analysis of fanfiction

Differences between canons and fanfiction

Beyond Canonical Texts: A Computational Analysis ofFanﬁction Smitha Milli Computer Science Division University of California, Berkeley [smilli@berkeley.edu](smilli@berkeley.edu) David Bamman School of Information University of California, Berkeley [dbamman@berkeley.edu](dbamman@berkeley.edu) Abstract While much computational work on ﬁction has focused on works in the literary canon, user-created fanﬁction presents a unique opportunity to study an ecosystem of literary production and consumption, embodying qualities both of large-scale literary data (55 billion tokens) and also a social network (with over 2 million users) . We present several empirical analyses of this data in order to illustrate the range of affordances it presents to research in NLP, computational social science and the digital humanities. We ﬁnd that fanﬁction deprioritizes main protagonists in comparison to canonical texts, has a statistically signiﬁcant difference in attention allocated to female characters, and offers a framework for developing models of reader reactions to stories. 1 Introduction The development of large-scale book collections—such as Project Gutenberg, Google Books, and the HathiTrust—has given rise to serious effort in the analysis and computational modeling of ﬁction (Mohammad, 2011; Elsner, 2012; Bamman et al., 2014; Jockers, 2015; Chaturvedi et al., 2015; Vala et al., 2015; Iyyer et al., 2016) . Of necessity, this work often reasons over historical texts that have been in print for decades, and where the only relationship between the author and the readers is mediated by the text itself. In this work, we present a computational analysis of a genre that deﬁnes an alternative relationship, blending aspects of literary production, consumption, and communication in a single, vibrant ecosystem: fanﬁction. Fanﬁction is fan-created ﬁction based on a previously existing, original work of literature. For clarity we will use the term CANON to refer to the original work on which a fanﬁction story is based (e.g. Harry Potter) and the term STORY to refer to a single fanauthored story for some canon. Although stories are based on an original canonical work and feature characters from the canon, fans frequently alter and reinterpret the canon—changing its setting, playing out an alternative ending, adding an original character, exploring a minor character more deeply, or modifying the relationships between characters (Barnes, 2015; Van Steenhuyse, 2011; Thomas, 2011) . In this work, we present an empirical analysis of this genre, and highlight several unique affordances this data presents for contemporary research in NLP, computational social science, and the digital humanities. Our work is the ﬁrst to apply computational methods to fanﬁction; in presenting this analysis, we hope to excite other work in this area. 2 Fanﬁction data Our data, collected between March–April 2016, originates [from](from fanfiction.net.1 In this)[ fanfiction.net](from fanfiction.net.1 In this)[.](from fanfiction.net.1 In this)[1](from fanfiction.net.1 In this)[ In this](from fanfiction.net.1 In this) data, AUTHORS publish stories serially (one chapter at a time); REVIEWERS comment on those chapters. A summary of data is presented in table 1 . The scale of this data is large for text; at 55 billion to- 1While terms of service prohibit our release of this data, tools to collect and process it can be found here: [http:](http:)//[github.com/smilli/fanfiction](github.com/smilli/fanfiction). Figure 1: Difference in percent character mentions between fanﬁction and canon for Pride and Prejudice (left) and Sherlock Holmes (right) . kens, it is over 50 times larger than the BookCorpus (Zhu et al., 2015) and over 10% the size of Google Books (at 468B tokens) . The dataset is predominantly written in English (88%), but also includes 317,011 stories in Spanish, 148,475 in French, 102,439 in Indonesian, and 73,575 in Portuguese. In total, 44 diff

FAQs

What makes fanfiction useful for computational research compared with canonical literature?

Fanfiction combines story creation, consumption, and communication in a social ecosystem, enabling analysis beyond what mediated only through published text allows.

How does the work compare character focus between fanfiction and canonical texts?

It analyzes character mentions by running BookNLP on canonical works and top fanfiction stories, then pairs characters across canon and fanfiction by name and measures changes in mention percentages.

What results are reported about gender and protagonist prominence in fanfiction?

Fanfiction shows a statistically significant difference in attention allocated to female characters and decreases mention percentages for main protagonists compared with canonical texts.

Aug 27, 2026

Upload by Bill Black