Skip to main content
Figure 1 | EPJ Data Science

Figure 1

From: Generalized word shift graphs: a method for visualizing and explaining pairwise comparisons between texts

Figure 1

Types of word contributions in word shift graphs. (A) Word contributions in basic word shift graphs, which are determined by the interaction between the signs of the difference between the word score and the reference score (\(+ / -\)) and the difference in relative frequencies (\(\uparrow / \downarrow \)) (see Sect. 3.1). For example, in sentiment analysis, a relatively positive word appearing more is indicated by a deep yellow bar to the right (+↑), while a relatively negative word appearing more is indicated by a deep blue bar to the left (−↑). (B) Word contributions in generalized word shift graphs, which additionally visualize the difference in word score (\(\triangle / \bigtriangledown \)) (see Sect. 3.2). If component contributions counteract one another then they are faded to emphasize the magnitude of the resulting contribution while retaining information about the detraction of one component from the other. For example, in sentiment analysis, if a relatively positive word is used more and its score is higher across contexts, then it is indicated by a deep yellow bar with an adjacent orange bar, both directed to the right (\(+\uparrow \triangle \)). If it is a relatively positive word that is used more but its score is lower, then the components counteract one another, indicated by a deep yellow bar to the right faded by the same amount as a subtractive purple bar to the left (\(+\uparrow \bigtriangledown \))

Back to article page