Generating Models of Recommendation Processes out of Annotated Ontologies

Creating content- and dialogue-based recommendation processes through manual adaptations requires a lot of time and effort. Therefore, automated generation of such processes is desirable. We present an approach for generating models of recommendation processes out of annotated ontologies. Such product ontologies have to be provided manually, but certain adaptations to them can be discovered from unstructured data (customer-generated content such as blog entries or customer feedback on products in the Web). They are given input for our approach, which applies semantic model-driven transformations to these ontologies for generating discourse-based models of recommendation processes on a high conceptual level first. These generated discourses essentially consist of questions and answers about those items annotated as important in the ontologies, and their possible sequences. From such a high-level model, transformation rules create a model of an operationalized recommendation process. This model also represents a so-called concrete user interface and consists of both the structure of the process and the course of events, which defines how customers may navigate through the process. From such models, an already given infrastructure can generate running processes including their final user interfaces, which have already been deployed successfully for real-world use.