Temporal Data Driven Naive Bayesian Text Classifier

Traditional text classifiers usually concern nothing about the producing time of samples, but many samples may involve seasonal features, which contain necessary prior information for classification. This paper firstly discovered the emporal relation between classes by means of chi-square test on a 2 * p dimensional contingency table for the data set of the mayor´s complain telephone texts, and then proposed a NB classifier for temporal data when driven by producing time of samples and utilized kernel regression for parameter estimation. The experiment showed a ignificantly improved classification performance.