During each comparison, the EFTG, IGPT, and CIQ schemes always have the same dataset.
Because the index is stored in memory, the first experiment tests the memory consumption of the EFTG, IGPT, and CIQ schemes.
On one hand, we remove the threshold from IGPT, which decreases memory consumption, and we add classification to the publish/subscribe system, which increases memory consumption.
When the number of keywords increases from 10 to 50, the memory consumption of IGPT and CIQ increases dramatically.
The runtime of EFTG is less than CIQ and IGPT. The EFTG scheme removes the threshold and classifies subscriptions.
Because IGPT and CIQ do not classify publications, each publication must be scored for all subscriptions by computing geo-textual similarities.
However, the matching numbers of IGPT and CIQ are likely to exceed k.
When the k value increases from 2 to 10, there are significantly fewer matches for the EFTG scheme than for IGPT and CIQ.
Therefore, the EFTG match time increases much more slowly than the match time for IGPT and CIQ.
IGPT and CIQ should compute the geo-textual similarity.
Meanwhile, the CIQ and IGPT update processes are almost unrelated to the k value, and their update times are therefore almost unaffected by the k value.