Text Classification Based on Enriched Vector Space Model

As one of the challenges to text classification can be indicated applying a model with the following characteristics: acceptable computational complexity of the model construction; dimension reduction of the vector space without significant decreasing of the classification performance. The present research aims to find a possible solution to mentioned problems. This paper proposes a model obtained by enrichment of the vector space model with the association relationships between words extracted from their co-occurrence in the text documents. For this purpose, the lift measure of association rules between word pairs is calculated. Experiments are conducted on Reuters-21578 dataset by using SVM classifier. The results confirm that applying the model improves the binominal and polynomial classification performance in comparison to the vector space model with respect to the F-measure even after word filtering, leading to a significant dimension reduction.