"What shall we do tonight?" Enhancing group recommendations with data mining techniques

(project, diploma, bachelor's, master's thesis)

Recommendation systems provide users with suggestions about products, movies, videos, pictures, books etc. They are very popular nowadays, e.g. Amazon, NetFlix, MovieLens.

Typically, recommendation systems focus on individual users, that is, they provide recommendations to a user about items that he/she might probably find interesting. So, a typical query could be "Which movie should I watch tonight?" or "What should I eat tonight?". To answer such queries we rely upon other users in the database that are similar to the given user.

However, there are contexts where the items to be suggested (e.g. movies, restaurants, concerts) are not intended for personal use but for a group of people, for example, a group of friends or family that is planning to watch a movie or visit a restaurant. In such cases, the above queries are rephrased as follows "Which movie should we watch tonight?", "What should we eat tonight?" Such kind of recommendations is known as group recommendations.

The goal of this project is to develop techniques for group recommendations. The straightforward solution that repeatedly scans the whole database of users is not efficient for an online application. So, our goal is to extract patterns from the database in an offline step and use these patterns afterward for online answering of group recommendations queries.

In this project, we will employ clustering for pattern extraction. For the experiments, we will a movie ratings dataset.

Requirements

• KDD/ Data Mining concepts (KDD I/ KDD II)
• Good programming skills (Java)
• Independent work

Contact
If you are interested in this topic and/or if you have further questions please contact: Eirini Ntoutsi (This is joint project with Kostas Stefanidis at the Norwegian University of Science and Technology)