Dissertation, Otto-Friedrich-Universität Bamberg, 2025

This work explores the integration of recommendations within spatial hypertext applications. Typically, spatial hypertext involves 2D interfaces where users can add, move, and spatially arrange objects. While hypertext usually refers to linking documents, in the case of spatial hypertext, these links are implicit, created through visual attributes rather than explicit connections. The position and arrangement of objects play a crucial role in this system. Such a system can be enhanced with system-generated recommendations: the user-created spatial structures are interpreted, and a knowledge base then provides recommendations that match the content and structure.

A previously unanswered question is how to present these recommendations effectively. This work focuses on developing, implementing, and evaluating methods for meaningfully embedding recommendations within existing visual structures. The goal is to make it easier for users to understand relationships and, thereby, enable more efficient and effective knowledge exploration. In this context, the term spatial "composition" is introduced, analogous to methods that interpret visual structures ("spatial parsing").

Five algorithms are developed and described as examples to achieve this goal. Their descriptions are chronological and highlight the problems encountered and their solutions. The algorithms are evaluated and compared based on various characteristics deemed important in the context of spatial composition. A user study examines the impact of integrating recommendations into spatial structures on effectiveness and efficiency, compared to a list-based presentation. In addition, the reasons behind these effects are investigated, considering future developments. Another user study explores the relationship between spatial proximity and the weighting of relationships between two objects. This is crucial because the proposed algorithms primarily use spatial proximity of objects to integrate recommendations in a way that is understandable for the user.

The results of the user studies demonstrate that, under the given conditions, a measurable improvement in efficiency is achieved. Further analysis indicates that this effect is indeed attributable to spatial composition. This is evident from various features observed throughout the testing sessions. Furthermore, it was found that the study participants perceive and use a linear relationship between spatial proximity and the weighting of object relationships to express such connections.

To demonstrate the practicality of the approaches developed, some of the algorithms have been implemented in research prototypes. These prototypes were primarily developed in collaboration with industry partners and cover a wide range of application domains. Spatial composition, combined with a knowledge base and a recommender system, facilitates the exploration of information by meaningfully presenting relationships to the user.