Fig. 3

We illustrate our approach by considering a hypothetical collection of abstracts belonging to two different meetings (far left). To quantify text overlap (same meeting) we use eTBLAST to assign similarity scores between pairs of abstracts withing those two meetings (middle left). Mutual overlap is possible; for example, there is a triplet U-Q-V of mutual text overlap in the hypothetical meeting 2 (middle left). Next, we randomly sample abstract pairs for review. Here, mutual overlap is possible but rare; for example, pair Q-U and pair Q-V in random sample 1 (middle right). Thus, 1 abstract could be compared with 2 or 3 possible overlaps, but rarely. Verified abstracts are used to estimate the rate of text similarity (within meeting)