The Historical Significance of Textual Distances

TL;DR

This paper investigates whether textual similarity measures accurately reflect cultural proximity by empirically comparing traditional and supervised learning-based methods in English fiction genres.

Contribution

It introduces new supervised learning strategies for measuring textual similarity anchored in social context, and compares them to existing methods.

Findings

Supervised learning methods better align with social measures.

Traditional cosine and topic vector similarities have limitations.

Empirical evidence supports the social relevance of new measures.

Abstract

Measuring similarity is a basic task in information retrieval, and now often a building-block for more complex arguments about cultural change. But do measures of textual similarity and distance really correspond to evidence about cultural proximity and differentiation? To explore that question empirically, this paper compares textual and social measures of the similarities between genres of English-language fiction. Existing measures of textual similarity (cosine similarity on tf-idf vectors or topic vectors) are also compared to new strategies that use supervised learning to anchor textual measurement in a social context.