Summary: The need for an accurate text classification model is increasing with the increase of the unstructured text such as online reviews, articles, and news. And one of the most critical steps in building such models is feature weighting. While TFIDF is the most popular method used of that purpose, it has some limitations that prevent the classification model from performing accurately in some cases. In this work we have two main contributions, first we are introducing an improvement version of TFIDF method using part of speech tagging that will overcome the limitations of the original TFIDF equation, second, we trained a new deep neural network with two hidden layers. The studies were conducted on four large datasets and the results were promising.
Keywords : Text mining, Feature weighting, Neural network, Part of speech, Natural language processing. |