Figure 2.
Graph construction: (1) The input text from Twitter. (2) Token extraction based on the Porter stemmer (Porter, 1980). (3) A graph is constructed where nodes are words and edges denote whether two words are juxtaposed in the text. (4) Scores reflect whether a node is reachable from the start of the input text. If more than 5 words reach a score of 10%, only the best 5 keywords are selected. If less than 3 keywords reach this threshold, the 3 best keywords are selected.