Automatic Classification of Web News: A Systematic Mapping Study

Abstract

The number of news articles published on the Web has had a dramatic increase. News websites are overwhelmed daily with articles, and their processing and classification is a challenge. Reading news from the web has become an important citizen’s information source, and its classification can show relevant information about social or cultural patterns on society. In this context, techniques that can automatically analyze and classify news articles are essential. In particular, data mining and machine learning techniques have been applied for the classification of web news, as they can detect structural patterns based on documents characteristics. Their use requires specialized text processing and summarizing techniques. The objective of this study is to characterize data mining and machine learning techniques used for the web news classification, the datasets used, and the evaluation metrics. We performed a systematic literature mapping of 51 primary studies published between 2000 and 2019. We found that the most used techniques fall into these paradigms: clustering, support vector machines and generative models. Also, 33 studies used online data extracted from Internet’s news web pages, while 25 downloaded a previously published dataset. The most common metric is the F-measure, with 25 reports. In summary, several data mining and machine learning techniques have been applied to the automatic classification of web news, showing some trends regarding the techniques, datasets, and metrics.