Improved Feature Weight Calculation Methods Based on Part-of-Speech in Text Classification

With the development of Information Technology and the increasing number of electronic documents, as a large-scale text information processing means, text classification attracts more and more attention on researchers. In order to obtain better performance in text classification works, two methods on improving the feature weight calculation by introducing the influence of part-of-speech are proposed, one is Single-Part-of-Speech (SPOS) and the other is Multi-Part-of-Speech (MPOS). Contrast experiments between the improved feature weight calculation methods and the original TF-IDF method are conducted. In terms of the improved approaches, the part-of-speech weights are optimized by the Particle Swarm Optimization algorithm. Besides, in order to prove that the improved methods are applicable, Reuters-21578 is used as the corpus in the experiment. The experiment results demonstrate that the improved feature weighting methods perform better than the original TF-IDF method by achieving higher precisions at different dimensions of feature space. In addition, MPOS method works more effectively than SPOS method. Through the in-depth analysis we can also find out that both noun and verb have certain extent of influence, but noun contributes relatively more to classification. Keywords—Text Classification; Part-of-Speech; Particle Swarm Optimization; Feature Weight;