Using a thematic model to enrich photo montages

Narrative systems attempt to present users with media collections that include some element of structure or story, however these collections can lack an authorial voice and seem bland as a result. In this paper we explore how themes could be used to enrich automatically generated narratives, and describe how a system which generated story selections in the form of photo montages was developed using a thematic model of narrative. This was achieved by selecting narrative atoms, in this case photographs, from a selection of images on a specific subject with relevance to a desired theme. Our pilot study shows that our thematic system selects images with greater relevance to desired titles, and that the positive impact of thematic selection increases when the images are presented together. We hope that our thematic work will inform others working on narrative systems, and will lead to richer automated narratives.