Classifying literary genres

Abstract

Classifying literary genres has always been methodologically confined to philological methods and what is commonly known as Vector Space Clustering (VSC). The problem has been exasperated with the widening gap between computational theory and traditional analysis of literary texts. Towards finding a solution to this problem, the current study utilizes a synergetic approach that brings together two established methods. First, a computational model of genre classification is drawn upon for identifying concept-based, rather than word-bound, topics, where the representation of texts is secured via the ‘bag of concepts’ (BOC) model as well as the sense-restricted knowledge and meaningful links holding between and among concepts; relatedly, the two model strands of explicit semantic analysis (ESA) and ConceptNet have enacted text classification. Second, a contextual lexical semantic approach (CRUSE, 1986, 2000) is employed so that the contextual variability of word meanings and concepts can be tackled within the confines of the target literary genres classified. The findings of present study have shown that the current composite approach of computational and semantic models has resulted in improved performance in classifying literary genres, especially with respect to delineating the links between each cluster’s document-members and generalizing about their unifying genre. Further implications have emerged from the present study, namely, the benefits reserved for digital libraries and the process of archiving, where literary-text classification has proven problematic to both users and readers in many cases.