Fig. 3

Analysis of the locality-based indicators (AbstractX and TITLE): Positions of entities can be measured in terms of words or sentences, as shown in (a) and (b) respectively. When compared with non-CAEs, CAEs are more likely to appear in titles of articles, as shown in the left-most parts of (a) and (b). On the other hand, when considering the abstracts of the articles, both CAEs and non-CAEs have somewhat uniform distributions at different positions, and moreover they have very similar distributions (see the two overlapping dashed lines). Therefore, TITLE works better than AbstractX in distinguishing CAEs from non-CAEs (see the solid line). However, TITLE has weaknesses as well, because most CAEs do not appear in the titles of articles (only 11.4% of CAEs appear in the titles)