Study of AI News Classification Model

Author(s): Mohammad Zaki, Shahil Ahamad, Fiza Khatoon and Anas Habib Zuberi

Abstract:

With the rampant expansion of the digital media platforms, there is an overburden in the amount of news content that is produced online on a daily basis. Such content has also been required to be effectively organized and categorized, in order to be retrieved, recommended to the user and to manage content. This paper introduces the design and implementation of an automated news classification system on both traditional machine learning and a deep learning approach. The dataset utilized in the paper is a collection of 10,000 labelled news stories gathered on Kaggle and classified into five different fields, specifically: Technology, Sports, Business, Education, and Entertainment of which there are 2,000 articles per field. Traditional models involve the use of the TF-IDF technique to extract the features to be used in their preprocessing pipeline (text cleaning, tokenization, stopword removal, and feature extraction). There were three classification algorithms that were used and compared: Naive Bayes, Logistic Regression, and Long Short-Term Memory (LSTM) networks. According to the experimental findings, the accuracy of the Naive Bayes classifier was 85, and the accuracy of the Logistic Regression was 87 and the LSTM model had a higher accuracy of 91. The results indicate that deep learning models are more effective in capturing contextual information as opposed to the traditional statistical techniques. The research points out the trade- offs in performance between computational and contextual based news classification tasks of multi-class news classification in multi-process computation versus contextual based computation.

keywords:
News Classification, Natural Language Processing (NLP), Machine Learning, Deep Learning, TF-IDF Feature Extraction, Long Short-Term Memory (LSTM),

Pages: 133-140     |    25 View     |    1 Download