Supporting People in Finding Information: Hybrid Recommender Systems and Goal-Based Structuring

The Internet has provided people with the possibility to easily publish and search for information. This resulted in an enormous amount of online available information, products and services that made it a challenge for people to find what is really interesting to them. This thesis addresses three solutions that can be used to support people in finding interesting items, based on the three main processes of a personalized information system: selecting, structuring and presenting information. For the selection of information, a framework is described for the development of hybrid recommender systems that switches between prediction techniques based on the most up-to-date knowledge about the user, other users, the item and other items. Experiments with the framework show that this increases the accuracy of interest predictions. To structure information, a method is investigated that makes it easier for people to find interesting information by structuring items according to the possible goals people have. Experiments with this goal-based structuring method show that goal-based structuring supports people in finding interesting information even more than recommendations. However, it takes people time to get used to goal-based structuring. Finally, it addresses those user interface aspects that are specific to recommender systems, namely presenting predictions, presenting explanations and providing feedback. A user-centered design approach shows that people prefer traditional ways of presenting predictions, that they believe explanations about predictions are important and that they prefer to not give explicit feedback but have a system derive their interests implicitly. However, if explicit feedback is required, it should be acquired in a manner similar to how predictions are presented.