Analyzing Motivating Texts For Modelling Human-Like Motivation Techniques In Emotionally Intelligent Dialogue Systems

In this paper, we present studies on human-like motivational strategies which eventually will allow us to implement motivational support in our general dialogue system. We conducted a study on user comments from a discussion platform Reddit and identified text features that make a comment motivating. We achieved around 0.88 accuracy on classifying comments as motivating or non-motivating using SVM and a shallow neural network. Our research is a first step for identifying computational features of a motivating piece of advice, which will subsequently be useful for implementing the ability to support the user by motivating him, imitating real human-to-human interactions.