The Role of Hubs in Cross-Lingual Supervised Document Retrieval

Abstract

Information retrieval in multi-lingual document repositories is of high importance in modern text mining applications. Analyzing textual data is, however, not without associated difficulties. Regardless of the particular choice of feature representation, textual data is high-dimensional in its nature and all inference is bound to be somewhat affected by the well known curse of dimensionality. In this paper, we have focused on one particular aspect of the dimensionality curse, known as hubness. Hubs emerge as influential points in the k-nearest neighbor (kNN) topology of the data. They have been shown to affect the similarity based methods in severely negative ways in high-dimensional data, interfering with both retrieval and classification. The issue of hubness in textual data has already been briefly addressed, but not in the context that we are presenting here, namely the multi-lingual retrieval setting. Our goal was to gain some insights into the cross-lingual hub structure and exploit it for improving the retrieval and classification performance. Our initial analysis has allowed us to devise a hubness-aware instance weighting scheme for canonical correlation analysis procedure which is used to construct the common semantic space that allows the cross-lingual document retrieval and classification. The experimental evaluation indicates that the proposed approach outperforms the baseline. This shows that the hubs can indeed be exploited for improving the robustness of textual feature representations.