Home
Consumers increasingly buy products through the Internet, but they lack assistance for searching the “right” product. Recommender systems address this problem by asking targeted questions. The success (or failure) of such a recommendation process is defined in terms of conversion rate or click-out rate. It is very difficult to predict improvements for given changes of the recommendation process, however, and manual changes are usually very expensive. Therefore within the SOFAR-project we propose automated adaptations of recommendation processes making use of semantic technology. Our approach makes Internet content accessible for searching products. In particular, ontologies are generated from Linked Open Data and unstructured information sources such as newsgroups, since ontologies are key to success. From their generic knowledge and related instances, high-level discourse models are automatically generated.
"Smart Assistant" online-product-advisors as example use-case-scenarios for feedback-driven adaptations in e-commerce recommender-systems.
These discourse models represent classes of potential dialogues between a customer and the recommender system and, in effect, a recommendation process. From these models, user interfaces for the end user are generated (semi-)automatically as well. After executing a discourse with its generated user interface for a period of time, a feedback component provides information about the usage of the system. This feedback leads to changes of the ontology, which in turn lead to changes of the discourse model. Consequently, the recommender process and its supporting user interface are changed as well. In addition, a human expert can influence this automated adaptation cycle. We will evaluate this approach through experiments with an existing recommender system owned by one of the proposers: Smart Assistant. Since this system is in successful real-world use, the evaluations do not have to be restricted to a laboratory.
The project is funded by the Austrian Research Promotion Agency (FFG) and the Federal Ministry for Transport, Innovation and Technology (BMVIT) under the FIT-IT “Semantic Systems” program (contract number 825061).
