The combination of text classification system

Abstract EN

Text classification is considered an important task for arranging text data under labels or categories.

Real life data varies in language, size, format, noise and area.

Consequently, these data need to be classified accurately due to their sensitivity and importance.

Despite there being different classifiers that could be used, any individual classifier is still affected by issues and does not produce the best result for all data.

Therefore, the researchers directed to use a system combination to enhance the prediction result.

In this paper, a combination system is used to avoid the variation in the classifiers' performance, by finding the best prediction among three classifiers (Naïve Bayes, Support Vector Machine, and K-Nearest Neighbors) before giving the final decision, to achieve the required reliability and the efficiency.

Four data collections are used in English and Arabic languages: 20-newsgroup, Reuters-21578, Watan-2004 and Khalaf-2018.

The proposed system produces better results than the individual classifiers.

The F1-scores for the proposed system were 95%, 93%, 94% and 92% for 20-newsgroup, Reuters-21578, Watan-2004 and Khalaf-2018, respectively