Timeline for Regression with variable containing multiple entries per observation - clustering right approach?
Current License: CC BY-SA 3.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 11, 2016 at 5:58 | comment | added | Has QUIT--Anony-Mousse | The algorithm may decide to ignore that one company belongs to both, too. It definitely won't care that a huma ginds "automotive" reasonable. But it might decide that car makers and airlines are the same, because they buy engines from rolls royce. | |
| Apr 11, 2016 at 1:35 | comment | added | JNWHH | For sure I would loose some details but grouping companies that have the tags "makes_cars" and "makes_trucks" under a more high-level tag resp. industry such as "automotive" (which may or may not exist) would be fine for me. | |
| Apr 10, 2016 at 14:44 | comment | added | Has QUIT--Anony-Mousse | Association rules aren't what you need. Frequent items are. But no; this will not give you 1 tag per company (which may not be a good idea anyway! why would every company be only allowed to have 1 tag?) Consider you have the tags "makes_cars" and "makes_trucks". There will be companies that do both, and companies that do only one of them. Reality is, patterns are not disjoint. It may be better to adjust your aims then. | |
| Apr 10, 2016 at 13:58 | comment | added | JNWHH | Thanks for the answer. r/ frequent itemset mining: I thought about this approach as well, but I am not sure whether this actually helps me achieving my goal. If I find certain association rules I might be able to combine certain tags and by that reducing complexity. However, I am not sure that this actually enables me to arrive at 25-100 "tags" with each company assigned to 1 at most. | |
| Apr 9, 2016 at 19:23 | history | answered | Has QUIT--Anony-Mousse | CC BY-SA 3.0 |