Various Aspects of Recommender Systems
Projects
The projects offered here are aligned with the lecture "Data Analysis and Query Languages" offered in the SS15, which covered some of the main RSs techniques. If you are interested in joining any of the below-listed projects or if you are willing to propose an idea contact any of the organizers.The specifications are password-protected. Contact any of the organizers to obtain the credentials.
- Recommending Plans for Visiting Touristic Attractions Exploiting GPS-Based Public Transportation System (part I)
Short description
More and more Public Transportation Systems (PTS), especially those in developing countries, are enhanced with real-time GPS-data sent from the transportation units. Timetables are not sufficient due to traffic jams, accidents, construction yards, bad weather, increased ridership or vehicle breakdowns. Input to this problem are a set of touristic attractions (given as GPS-locations) and the real-time GPS-data sent from transport units. The goal is to recommend a set of touristic attractions to be visited in a given time along with a journey plan. The recommender has to be implemented as a smartphone application.
Specification: Download specification 
Recommended number of participants: 2
Status: assigned (2/2)
- Collaborative filtering in a cluster computing platform
Short description
The goal of the project is to integrate one of the most widely adopted algorithms for recommendations, collaborative filtering (CF), with the power of cluster computing platforms like SPARK, a big data processing engine. SPARK makes it possible to run K-neighbors, a subtask of CF, efficiently by distributing the load of computation over the nodes. This horizontal scalable platform allows this component to run in-memory. An evaluation will assess the performance and the quality of the recommendations obtained by the enhanced approach.
Specification: Download specification
Recommended number of participants: 2
Status: assigned (2/2)
- Enhanced user profile for content-based recommender systems
Short description
A content-based recommender system (CBRS) typically requires a user profile (UP) that represents its interest for specific attributes of the items. For instance, in the case of movies those attributes could be the directors, producers, cast, genres, etc. The CBRS finds the best matches between a UP and available items not yet consumed by that user. The profiler learner is therefore an important component in the system. If the UP is not able to properly represent the taste of a user, the recommendations will reflect the lack of quality. The goal of the project is implement state-of-the-art techniques for building the profile and to explore different approaches to enhance it.
Specification: Download specification
Recommended number of participants: 2
Status: assigned (2/2)
- Enhancing recommendation of publications with taxonomies
Short description: PubRec is a recommender system that helps researchers find interesting and related scientific publications. Considering the huge number of rapidly published papers, researchers are overwhelmed with the countless number of potentialy important papers. PubRec analyses the researcher previous publications and extracts a researcher profile that represents his/her research interest. In this project we aim to extend PubRec in the following directions:
- Incorporating domain related taxonomy within a previously developed content-based approach for recommending research publications
- Developing a web-based interface for PubRec
Recommended number of participants: 2
Status: assigned (1/2)