Modified Approach of Multinomial Naïve Bayes for Text Document Classification

This work proposes a text classification using modified approach of Multinomial Naïve Bayes for justifying and identifying the documents into a particular category. Due to the exploration of the textual information from the electronic digital documents as well as World Wide Web. Naïve Bayes theorem is effective for classification of text documents into the predefined categories by means of the probabilistic values. However, its performance is repetitively inadequate by inappropriate feature selection. The aim of this paper is to propose a method that will improve the classification accuracy decision. In addition a new feature selection method for text document classification in machine learning is also proposed. In machine learning the training set is generated for testing the documents. Scoring method is used to enhance the efficiency of both classifications with a relevance to accuracy and performance. Keyword: Text classification, Naïve Bayes, Feature selection. Introduction With the increasing availability of electronic documents and rapid growth of the World Wide Web and data in digital format, the task of automatic document classification is important for organization. Proper classification of electronic documents, online news, blogs, e-mails and digital libraries requires Text Mining, Machine learning and natural language processing techniques to extract required knowledge information. Text mining makes an attempt to discover interesting information and knowledge from unstructured documents. The important task is to develop the automatic classifier to maximize the accuracy and efficiency to classify the existing and incoming documents. Voluminous information of an organization is stored in an unstructured form of reports messages, news and email [1]. However, data mining deals with structured data, whereas text presents special characteristics and is unstructured. The important task is how these documented data can be properly retrieved, presented and classified it is difficult to machine, so there has been a growing interest in this area of research [2]. Extraction, integration and classification of text documents from different sources and knowledge information discovery which finds features from available documents are important. In data mining, Machine learning is often used for Prediction or Classification. Classification involves finding rule that partitions the data into disjoint groups. The input for the classification is the training data set, whose class labels are already known. Classifications analyze the training data set and construct a model based on the class label. The goal of classification is to build a set of models that can correctly predict the class of the different objects. Machine learning is an area of artificial intelligence concerned with the development of techniques which allow computers to "learn". More specifically, machine learning is a method for creating computer programs by the analysis of data sets since machine learning study the analysis of data. The challenging task is of text classification performance, because many problems are due to high dimensionality of feature space and unordered collection of words in text documents. Various machine learning algorithms available and utilize in document classification. Naïve Bayes has been one of the popular machine learning algorithm because of its simplicity. [2][3][4] easy to implement and draws better accuracy in large datasets[5]. Naïve Bayes classifier performing well in classification task where the probability is calculated by the Naïve Bayes independent assumption [6] [7]. The paper mainly focuses on reducing the number of features class dependent by using the text document feature selection with new feature scoring method and using proposed feature selection method implement Modified approach of Multinomial Naïve Bayes classification model to classify testing text documents. Thousands of term word occurs in the text document, so it is important to reduce the dimensionality of feature using feature selection process [8], to resolve this problem many feature evaluation metrics have been explored such as X2 Statistics (CHI), Information Gain (IG), mutual information, term strength, document frequency, Term Frequency Inverse Document Frequency. With the help of these approaches it is possible to reduce the high dimensionality of features. Proposed feature scoring metrics to select the feature is the most effective method to reduce the dimensionality of feature and improve the efficiency and accuracy of classifier. In this approach document preprocessing is also important to reduce the complexity and high dimensionality of features occurs in the text document. Vol 6 • Number 2 April Sep 2015 pp. 196-200 Impact Factor: 2.5 Available at www.csjournals.com DOI: 10.090592/IJCSC.2015.614 Page | 197 Feature selection Approaches Feature selection helps in the problem of text classification to improve efficiency and accuracy. In our approach we are examining different feature selection methods and then will find wheather our proposed method is effective to other studied method. A. TF (Term Frequency) Term frequency in the given document is simply the number of times a given term appears in that document. TF used to measure the importance of item in a document, the number of occurrences of each term in the document. Every document is described as a vector consisting of words. Importance of the term „t‟ within the particular document with „ni‟ being the number of occurrences of the considered term and the denominator is the number of occurrences of all terms.