Classification of Andrea Camilleri’s works using bottomup and top-down approaches

This study aims to classify Andrea Camilleri’s works into groups based on their stylistic similarities, employing various approaches. The available corpus consists of 70 works published by Sellerio spanning approximately 40 years (1978–2024). First, an exploratory approach was taken, using unsupervised methods (bottom-up classification), such as cluster analysis and multidimensional scaling. This revealed some notable trends, particularly in a collection of short stories and some standalone works. For example, the standalone novel ‘La rizzagliata’ bears similarities to the Montalbano novels, probably because it mentions the Inspector, despite not being part of the series. Thus, while the reasons for the anomalies found in the exploratory analyses are clear for some works, others require further investigation. To extrapolate additional clues from these results, we performed a supervised classification using information from the special issue of Quaderni Camilleriani published in 2021. This publication identified three main categories: ‘Inspector Montalbano’; ‘Historical and Civic Works’; and ‘Fiction of Various Kinds’. We added an additional residual class to this categorisation, consisting of plays, dialogues and letters — writings that were not included in the three main categories of Quaderni Camilleriani. Supervised (top-down) classification using a Support Vector Machine with a Radial Basis Kernel Function produced highly accurate results, particularly through the most frequent words and above all through the axes of a Correspondence Analysis. Exclusive use of functional or ‘empty’ words does not achieve the best result, as might be expected given the presence of several neologisms (idiolect) typical of Camilleri. Overall, the results of the analysis demonstrate the potential of distant reading and classification methods, while leaving ample room for qualitative analysis of the writing style of this prolific and complex author.