kcarnold
commented
13 years ago [from sgt101 on Launchpad] Related to this it would be very useful to put the info on canonical documents into some structured file like .xls or .csv separate from the general concepts file that is in results now.
I would like to see the following records produced :
filename.txt,polarity_value, concept 1, concept 2, concept 3
The concepts should not be word occurences but generalised terms from the analogy space...
When a canonical document is selected, the info pane should show a list of commonly-occuring concepts in the documents that are most similar to it.
These should probably be pre-SVD counts, not post-SVD similarity scores. For example, counting the number of times the words "chinese" and "thai" occur, weighted by which documents they are in, but not weighted by anything involving the "chinese" and "thai" concept vectors themselves. This would reassure users that the data reflects reality, even if the SVD comes out kind of weird.
Probably the best way to report these values would be as percentages.
[This bug transferred from Launchpad]