Naslov
Proceedings of the ECML-PKDD 2011 Workshop on Discovery Challenge

Urednik/ci
Šmuc, Tomislav ; Antulov-Fantulin, Nino ; Morzy, Mikołaj

Vrsta, podvrsta i kategorija knjige
Uredničke knjige, zbornik, znanstvena

Izdavač
Institut Ruđer Bošković

Grad
Zagreb

Godina
2011

Stranica
109

ISBN
978-953-6690-89-3

Ključne riječi
Machine Learning; VideoLectures.Net; Recommender Systems; Data Mining

Sažetak
The 2011 ECML-PKDD Discovery Challenge deals with the learning problems from the domain of recommender systems. Datasets and problems designed by the organizers of this Challenge, originate from the VideoLectures.Net site, a free and open access multimedia repository of video lectures, mainly of research and educational character. The lectures are given by distinguished scholars and scientists at the most important and prominent events like conferences, summer schools, workshops and science promotional events from many fields of science. The Challenge was organized with multiple aims in mind: to improve the current websites recommender system, discover new algorithms or computational workflows and provide new dataset for the research community. It encompassed two tasks: first one related to new-user/new-item recommendation problem, and the second task in which ”normal mode”, click-stream based recommendation is simulated. Dataset for the challenge is somewhat specific as it does not include any explicit nor implicit user preference data. Instead, implicit profiles embodied in viewing sequences have been transformed into a graph of lecture co-viewing frequencies and pooled viewing sequences. The data also includes content related information: topic taxonomy, lecture titles, descriptions and slide titles, authors’ data, institutions, lecture events and timestamps. The dataset (including the leaderboard and the test set) will remain publicly available for experimentation after the end of the challenge. Over 300 teams registered for the challenge, resulting in more than 2000 submitted results for the evaluation from 62/22 active teams for task 1 and task 2, respectively. The teams approached the tasks with diverse algorithms and in several cases novel feature construction approaches.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo

POVEZANOST RADA