An ensemble based NLP feature assessment in binary classification

Text feature selection plays an important role in text mining. Terms are the key players in document representation. The document representation can help application in following areas-indexing, summarization, classification, clustering and filtering. Text instances come with a challenge of high dimensional feature space and using such features can be extremely useful in text analysis. Hence it is important to extract important terms from a document. In this paper, we examine the impact of NLP features (stop words, stemmer and combination of both) on predictive performance of base classifiers and ensembles of Naive Bayesian category. We selected different category of base classifier like NB, SVM, KNN and J48 as these are frequently used by the researchers in text mining. IMBD movie review dataset is used as a standard dataset for experimental work. We prepared ensembles of Naive Bayesian with base classifiers and found ensemble gives better performance over the base classifiers with entire NLP categorical dataset. Ensemble of NB with SVM out performed among other ensembles with different categorical dataset.