An exploration of learning to link with wikipedia: Features, methods and training collection

From WikiPapers
Jump to: navigation, search

An exploration of learning to link with wikipedia: Features, methods and training collection is a 2010 conference paper written in English by He J., De Rijke M. and published in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics).

[edit] Abstract

We describe our participation in the Link-the-Wiki track at INEX 2009. We apply machine learning methods to the anchor-to-best-entry-point task and explore the impact of the following aspects of our approaches: features, learning methods as well as the collection used for training the models. We find that a learning to rank-based approach and a binary classification approach do not differ a lot. The new Wikipedia collection which is of larger size and which has more links than the collection previously used, provides better training material for learning our models. In addition, a heuristic run which combines the two intuitively most useful features outperforms machine learning based runs, which suggests that a further analysis and selection of features is necessary.

[edit] References

This section requires expansion. Please, help!

Cited by

Probably, this publication is cited by others, but there are no articles available for them in WikiPapers.