Description

Supervision: Gjergji Kasneci, Maximilian Jenders

This seminar will give an overview of state-of-the-art recommendation techniques and methods for dealing with large and sparse user-item matrices. It will be held in English and will consist of three parts.

In the first part of the seminar the students will be introduced to the above mentioned techniques. The main material will consist of work (in form of scientific papers).

In the second part, the students will form groups; each group will be given the opportunity to implement basic data structures for scalable recommendation techniques. 

Finally, in the third part, the students will implement recommendation algorithms that build on such data structures. The algorithms will be evaluated on a large-scale dataset extracted from various social news sites.

At the end of the semester, each team will deliver a brief 2-4 page summary of the pursued project.

The seminar will start with the beginning of the 2013 summer semester on Mo., April 8th.

In the first two weeks, the students will have the opportunity to study related research articles, which will be discussed in the seminar meetings (Mondays and Wednesdays).

By the end of the second week the students will form two person teams; each of the teams should decide which topic it would like to work on. Once the students have selected their favorite topics, there will be only weekly or biweekly feedback and progress meetings with individual teams.

There will also be one intermediate and an end presentation from each group.

The grading will approximately follow the distribution below:

• 25% for presentations
• 40% implementation
• 20% project report
• 15% participation in feedback and progress meetings

Work in this seminar will be rewarded with 6 LP.

1. Koren et al., IEEE 2009
2. Y. Koren, TKDD 2010
3. A. Gionis et al., VLDB 1999
4. D. Ravichandran et al., ACL 2005
5. Paulevé et al., Pattern Recognition Letters 2010

Slides from the intermediate presentation:

Group 1 (Maximilian Schneider, Martin Kreichgauer): Google-Docs

Group 2 (Thorben Lindhauer, Sebastian Kruse): HPI internal

Group 3 (Nils Rethmeier, Fabian Tschirnitz): Google-Docs

Group 4 (Robin Schreiber, Toni Mattis): HPI internal

Group 5 (Wiradarma Pratama, Uwe Hartmann): HPI internal