Discriminative category matching: efficient text classification for huge document collections

With the rapid growth of textual information available on the Internet, having a good model for classifying and managing documents automatically is undoubtedly important. When more documents are archived, new terms, new concepts and concept-drift will frequently appear Without a doubt, updating the classification model frequently, rather than using the old model for a very long period is absolutely essential. Here, the challenges are: a) obtain a high accuracy classification model; b) consume low computational time for both model training and operation; and c) occupy low storage space. However, none of the existing classification approaches could achieve all of these requirements. In this paper, we propose a novel text classification approach, called discriminative category matching, which could achieve all of the stated characteristics. Extensive experiments using two benchmarks and a large real-life collection are conducted. The encouraging results indicated that our approach is highly feasible.