Closed AlexanderKidd closed 5 years ago
The test page, factoidl_tests_page.html, took about 1m15s, which should really be closer to the max time it takes for a long page of 500-1000 factoids.
Will hopefully try to wrap up a solution tomorrow by checking bottlenecks and perhaps caching repeated processing like factoid normalization. But the other parts of the algorithm may have to be held off until a minor update to Socrates since the progress so far is worthy of a tag if it can perform reasonably well with speed and accuracy.
On second look, it isn't that bad. I will proceed with caution in the coreference resolution part of the algorithm, and see if it is too much to handle, and do the same with synonym textual entailment.
...Not sure if implementing coreference resolution or synonym textual entailment is viable if it takes over 2 minutes for small-medium responses and 10 minutes or more for the more lengthy ones. Could be a bug with the web workers getting overflowed by multiple tabs' data but they should be one per page...