Today I have been notified that all our three submissions to the ECIR 2009 conference have been accepted for publication.
In the paper “Active Learning Strategies for Multi-Label Text Classification” we investigate the problem of performing active learning in a multi-label context.
A typical assumption made in works on active learning is that the human annotator validates each category separately. This assumption hardly applies in practice, whereas is more convenient to ask to the annotator to validate a document on all the categories at the same time.
We explore various strategies for combining the confidence scores assigned to documents on all the categories, producing a single document ranking, in order to perform efficient and effective active learning on multi-label data.
In the paper “Multi-Facet Rating of Product Reviews” we face the problem of deriving rating score about specific aspect of reviews.
The main part of the paper is focused on comparing the effect of adding pattern-matching-based and sentiment-lexicon-based features to the documents representation, in order to improve the efficacy of the rating system.
We also discuss about using a macro-averaged evaluation measure for ordinal regression tasks on non-uniformly distributed data.
In the poster paper “Encoding Ordinal Features into Binary Features for Text Classification” we propose a method that enables learning algorithms based on binary features to make use of the information available from ordinal features, i.e., features sorted on an ordinal scale. In the specific case we study how to encode term frequency information as binary features in documents representation.
See you in Toulouse.