II.2.4 Machine Learning and Time series analysis
Once independent variables, either as discrete values, or as time series are established (see section II.2.2), and dependent variables are available as ground truth (see section II.2.3), a machine learning system can be built. Depending on the nature of the data, different machine learning algorithms are most suitable. For discrete values such as actor-level betweenness values, machine learning algorithms such as Bayesian classifiers, decision trees, random forests, and SVM work well. Such a model can be used for example to predict customer satisfaction based on “honest signals” calculated from e-mail for each employee of a company. For natural language processing, recurrent neural networks (RNN) such as long short-term memory (LSTM) are well suited, as they can handle word order in sentences. LSTM is also well suited for time series analysis, for example predicting tomorrow’s stock price based on a time series of the last seven days of the Twitter activity of the stock. Convolutional Neural Networks (CNN) are multilayer neural networks that are well suited for recognizing images. They are thus predominantly used for analyzing emotions from facial expressions and body posture, be it of humans or of animals. In the subsequent chapters we will look at all of these use cases in more detail.