Text Classification Model Based on Semantic Pattern Vector Space

摘要

The existing text classification systems are mostly based on the bag of words vector space model (BOW-VSM) and classify relevant documents by representing them as the bag of words vector. The BOW-VSM-based text classification systems, in spite of their merits of classification velocity and efficiency, have failed to represent the complete semantics contained in the document contents and have led to the higher vector space dimensionality and the lower classification performance. Generally, a sentence is the smallest semantic unit of representing an integrated semantics. So we take the sentences of documents as the smallest semantic disposal unit and construct the sentence-level semantic patterns, consequently, forming the semantic pattern vector space model (SP-VSM) of documents. Then on which we base to design and implement an automatic text classification system. Finally, we cany out a contrast experiment to the classification performance of two kinds of classifiers with two different vector space models. The experimental results show that the classification performance of the SP-VSM-based text classification system has a greater improvement than that of the BOW-VSM-based.