As was discussed in Chapter 3, Mining the 20 Newsgroups Dataset with Clustering and Topic Modeling Algorithms, unsupervised learning, including clustering and topic modeling, can be applied to text data. We will continue to see how supervised learning, specifically classification, is used in the text domain.

In fact, classification has been widely used in text analysis and news analytics. For instance, classification algorithms are used to identify news sentiment, positive or negative in a binary case, or positive, neutral, or negative in a multiclass classification case. News sentiment analysis provides a significant signal to trading in the stock market.

Another example that comes to mind is news topic classification, where classes may or may not be mutually exclusive. In the newsgroup example that we just worked on, classes are mutually exclusive (despite slight overlapping), such as computer graphics, motorcycles, baseball, hockey, space, and religion. We will demonstrate how to use machine learning algorithms to solve such multiclass classification problems in Chapter 5, Classifying Newsgroup Topic with Support Vector Machine. It is, however, good to realize that a news article is occasionally assigned multiple categories, where, properly speaking, multi-label classification is more suitable. For example, an article about the Olympic Games may be labeled as both sports and politics if there is an unexpected political involvement.

Finally, a text classification application that is perhaps difficult to realize is named-entity recognition (NER). Named entities are phrases of definitive categories, such as names of persons, companies, geographic locations, dates and times, quantities, and monetary values. NER is an important subtask of information extraction, to seek and identify such entities. For example, we can conduct NER on the following sentence: SpaceX[Organization], a California[Location]-based company founded by a famous tech entrepreneur Elon Musk[Person], announced that it would manufacture the next-generation, 9[Quantity]-meter-diameter launch vehicle and spaceship for the first orbital flight in 2020[Date].

To solve these problems, researchers have developed many power classification algorithms, among which Naïve Bayes and support vector machine (SVM) models are often used for text classification. In the following sections, we will cover the mechanics of Naïve Bayes and its in-depth implementation, along with other important concepts, including classifier tuning and classification performance evaluation.