International Conference on Instrumentation, Measurement, Circuits and Systems (ICIMCS 2011)
253 A New Input Representation for Multi-Label Text Classification
Download citation file:
- Ris (Zotero)
- Reference Manager
Automatic text classification is the task of assigning unseen documents to a predefined set of classes or categories. Text Representation for classification has been traditionally approached with tf.idf due to its simplicity and good performance. Multi-label automatic text classification has been traditionally tackled in the literature either by transforming the problem to apply binary techniques or by adapting binary algorithms to work with multiple labels. We present tf.rfl, a novel text representation for the multi-label classification approach. Our proposal focuses on modifying the data set input to the algorithm, differentiating the input by the label to evaluate. Performance of tf.rfl was tested with a known benchmark and compared to alternative techniques. The results show improvement compared to alternative approaches in terms of Hamming Loss.