Suggestion Models to Support Personalized Information Filtering

During the last few decades, a lot of information has become available on the web. It is evident that with this large amount of available data, it is difficult for people to find what they are looking for, they can feel overloaded and it can become complex for them to solve their search task. To overcome these problems, search engines and recommender systems have become an important part of most of the online services available nowadays, since they are able to assist the users in the process of retrieving relevant information on the web and they help the users to discover new items that they were not aware of before. In this thesis, we analyze different aspects in the field of information retrieval, recommender systems, and human-computer interaction in order to improve the intelligence of a recommender system with the final goal of offering users a mixed-initiative model that helps them to explore the retrieved and suggested information in order to satisfy their information needs. From the algorithmic perspective, we describe different recommendation models that leverage several types of information at different granularity levels. Specifically, we analyze the impact of ratings, search queries, topics and categories and trust information on item recommendation by employing them in distinct recommendation models. However, we are not only interested in improving the performance of the recommender systems but we are also interested in investigating the use of an appropriate user interface that allows users to inspect and interact with the retrieved information. During the past few years, it emerged the need of offering to the user a mixed-initiative interactive model that mixes the intelligence coming from the recommender system with the possibility for the users to tune and interact with the retrieved information. Thus, from the human-computer interaction perspective, we investigate the use of a set of widgets to help the users to explore the retrieved information in a map-based web application. In the future, we plan to use the insights that we collected from the results of the works presented in this thesis to build a hybrid recommender system to improve the recommender system intelligence and to integrate it inside a web application in order to offer to the users a mixed-initiative interaction model.