Information Theoretic Text Classification

The assignment of natural language texts to two or more predefined categories based on their contents, is an important component in many information organization and management tasks. This paper presents an information theoretic approach for text classification problem that we call it ITTC. Here, we prove that ITTC is theoretically equivalent to Bayesian classifier. However, when classification task is performed over dynamic or noisy data, or when the training data do not represent all probable cases, ITTC outperforms Bayesian classifier. We also show that the complexity of ITTC over test set grows linearly by the size of input data .We use some news groups, to evaluate the superior performance of our approach.

کلیدواژه ها:

Bayesian classifier ، Entropy ، Markov chain ، text classification