This study aims to investigate and visualize sensationalism configurations in nineteenth-century Romanian serial novels by employing Sentiment and Emotion Analysis. According to literary history, one of the defining strategies of such texts is sustaining the reader’s curiosity across successive feuilletons, often through sensational, melodramatic, and cliffhanger endings. In order to test these remarks, we aim to track the fluctuation of the reader’s affective state and emotional responses throughout the reading process, in correlation with the novel’s serialized structure. Our research questions are the following: 1. Can we quantify and visualize sensationalism patterns based on the reader’s affective response? 2. Based on dominant emotions in feuilleton transitions, are there different types of sensationalism? To this end, our analysis centres on a small collection of popular novels, whose particularity is that they have been digitally edited in the Pop-Lite Corpus according to their original feuilleton structure. Following a strict protocol and preceded by extensive discussion, the annotation has been carried out manually. The study is an extension of a previous experiment that showed both a consistent sensationalism pattern based on sentiment shifts and different emotional strategies in two sets of feuilletons belonging to different authors. Previous insights will be validated through the analysis of a broader dataset, with the overarching objective of proposing a quantitative definition of sensationalism in nineteenth-century Romanian serial fiction.

Medieval Latin alchemy, still fresh literary tradition, constituted a dynamic confluence of multiple literary genres. Drawing on broader contemporary narrative traditions while simultaneously cultivated a semi-fluid genre typology shaped by utilitarianism and esoteric strategies. Narrative texts, for example, contained not only narration itself but also densely symbolic allegories and theoretical excursions, alongside practical recipes and descriptions of alchemical praxis and laboratory instruments. Within certain alchemical traditions, particularly within florilegia (excerptive anthologies), the distinction between narrative fiction and instructive recipe literature could become remarkably thin. Consequently, the tradition of alchemical writing continues to pose significant interpretative challenges for contemporary scholars.

This study draws on a corpus of over fifty digitised medieval Latin alchemical texts and proposes a taxonomy of alchemical genres through the computational analysis of their sectional composition. Supervised machine learning classification techniques, model genre differentiation based on the manually labelled textual sections. These features are based on the level of their narrativity, descriptivity and use of formal structures (e.g. recipes). Feature extraction enables the identification of recurrent compositional patterns of identified “building blocks” across texts and their variants.

Training, tuning and evaluating of classification models on annotated data allows for the systematic characterisation of literary genre categories at both the macro-structural level of Latin alchemy and in terms of intra-textual dynamics. The results illuminate strategies of textual organisation, degrees of genre stability, and the distribution of functional elements across genres of Latin alchemical genres. In addition to offering new insights at the level of textual analysis, the project contributes to the modelling of a complex literary tradition.

Literature curriculum in Greek primary and secondary education is focused on state-assigned anthologies. Texts in those anthologies are organised in thematic groups until level 8. There is little research on how texts semantically align with those thematic groups or how themes are distributed across educational levels or in the whole thematic profile of Greek literature curriculum.

This project addresses this gap by applying computational methods across literature curriculum books. AI based methods such as sentence-embeddings, BERTopic and zero-shot assignment are used. Multiple pretrained multi-language sentence transformers embedding models are used throughout.

Initially, a way of creating stable target thematic groups from the pre-defined themes is constructed. Using sentence embeddings and BERTopic with a range of parameters and models, a set of more than 3000 groupings were formed. The resulting groups are the ones that appear together more often, using co-occurrence.

In the main part, zero-shot similarity-based assignment was used to assign texts to thematic groups centroids using distance metrics. To identify both strong and subtle themes, hard similarity (most similar text to thematic group) as well as soft similarity (distance to each thematic group) were used. Using the above assignments comparison of the assigned theme and the pre-existing theme can be performed, and the full corpus can be analysed.

The study produces results grounded in a systematic, transparent, and reproducible analytical framework, examining thematic patterns and thematic structure across the literature curriculum. The findings reveal patterns of thematic distribution and thematic overlap across all educational levels. The study does not replace human interpretation or qualitative literary analysis; rather, it complements them.

This research contributes to curriculum studies in literature by introducing large-scale AI-based semantic analysis as a neutral, computational perspective on the full corpus of school literature textbooks, offering an unbiased, reproducible approach to examining thematic structure through sentence embeddings.