An Automated Pipeline for Character and Relationship Extraction from Readers Literary Book Reviews on Goodreads.com

Abstract

Reader reviews of literary fiction on social media, especially those in persistent, dedicated forums, create and are in turn driven by underlying narrative frameworks. In their comments about a novel, readers generally include only a subset of characters and their relationships, thus offering a limited perspective on that work. Yet in aggregate, these reviews capture an underlying narrative framework comprised of different actants (people, places, things), their roles, and interactions that we label the "consensus narrative framework". We represent this framework in the form of an actantrelationship story graph. Extracting this graph is a challenging computational problem, which we pose as a latent graphical model estimation problem. Posts and reviews are viewed as samples of sub graphs/networks of the hidden narrative framework. Inspired by the qualitative narrative theory of Greimas, we formulate a graphical generative Machine Learning (ML) model where nodes represent actants, and multi-edges and self-loops among nodes capture context-specific relationships. We develop a pipeline of interlocking automated methods to extract key actants and their relationships, and apply it to thousands of reviews and comments posted on Goodreads.com. We manually derive the ground truth narrative framework from SparkNotes, and then use word embedding tools to compare relationships in ground truth networks with our extracted networks. We find that our automated methodology generates highly accurate consensus narrative frameworks: for our four target novels, with approximately 2900 reviews per novel, we report average coverage/recall of important relationships of >80% and an average edge detection rate of >89%. These extracted narrative frameworks can generate insight into how people (or Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s).

Conspiracy in the Time of Corona: Automatic detection of Emerging Covid-19 Conspiracy Theories in Social Media and the News

“…With minimal supervision, a few actant mentions are grouped together including, [trump, donald] : donald trump, [coronavirus, covid19, virus] : coronavirus and [alex, jones] : alex jones. While such groupings are not strictly required and could be done more systematically (see [59]), this actantgrouping enhances the co-occurrence graph by reducing the sparsity of the adjacency matrix representing subject-object interaction.…”

Modelling social readers: novel tools for addressing reception from online book reviews

“…This model has the advantage that it can show multiple, at times competing, claims to the underlying storyline (or storylines) of the target work. In previous work, we introduced a pipeline addressing two important tasks that are instrumental in constructing this network representation: entity mention grouping (EMG) and inter-actant relationship clustering (IARC) [ 60 ], which we summarize below.…”

“…In earlier work, relationship extraction along with the two-step process (EMG, IARC) to condense the relationship tuple space was applied to the corpus of reader reviews for four works of literary fiction: Of Mice and Men , To Kill a Mockingbird , The Hobbit and Frankenstein [ 60 ]. The resulting aggregated networks, which we label ‘narrative frameworks’, represent the broad consensus across all the reviews of the story network, with each node in the network representing a character and each directed edge representing a relationship between a pair of characters.…”

“…The IARC task, on the other hand, is designed to aggregate relationship phrases generated as output from a relationship extraction module. A relationship tuple consists of a subject mention, relationship phrase and object mention (sfalse^,rfalse^,ofalse^) that is directly extracted from a review and thus contains partial information about the structure of the underlying narrative model [60]. For every ordered pair of characters { e i , e j }, we obtain an aggregated set of relationship phrases from the reviews, R^ij, that connects e i to e j .…”

Exploring Brazilian Cultural Identity Through Reading Preferences

“…In such a context, Goodreads is one of the leading social networks for book readers. Current research on such a platform includes using book reading behavior to predict best-sellers [Maity et al 2017], measuring book impact by user reviews [Wang et al 2019], and using such reviews to extract meaningful characters and their relationships [Shahsavari et al 2020]. Data extracted from Goodreads is also used to understand cultural differences between countries.…”