Shift adjacency networks and emotional transition networks in film scripts: a gender-stratified structural analysis

A narrative admits multiple graph representations, and distinct constructions preserve non-equivalent topological information. On 1,655 film scripts, we investigated whether structural descriptors carry a signal on reception indicators and whether this relationship depends on gender. We defined three representations: the undirected network of adjacency of speech turns, the directed network of transitions between discrete affective states per scene, and the time series of the protagonist's feeling as a baseline. About them, we extracted 18 descriptors and evaluated their association with IMDb score and box office by Spearman's partial correlation, controlling for year and budget. Candidate associations identified in training (films up to 2005) were examined only once in temporal hold-out (n=146 post-2005), with criteria set a priori. Together, the descriptors do not add variance over the baseline; Stratified by gender, the analysis reveals structure, and most associations are confirmed. In this study, stratifying by gender was necessary to detect such signs.