Learning an optimal distance metric in a linguistic vector space

Many natural language processing still depend on the Euclidean distance function between the two feature vectors, but it has severe defects as to feature weightings and feature correlations. In this paper we propose an optimal metric distance function that can be used as an alternative to the Euclidean distance, accommodating the two problems at the same time. This metric is optimal in the sense of global quadratic minimization, and can be obtained from the clusters in the training data in a supervised fashion. We confirmed the effect of the proposed metric by the sentence retrieval, document retrieval, and the K-means clustering of general vectorial data.