My idea here is to combine two types of info for each sender: personal and global.
Personal would be a classifier trained only on this particular sender. Global would be trained with no consideration for the sender.
It would allow for cases when a subordinate calls his boss "Sir", but a partner calls him by his name for example.
We could weight the personal classifier by how much training examples we had for it.
My idea here is to combine two types of info for each sender: personal and global. Personal would be a classifier trained only on this particular sender. Global would be trained with no consideration for the sender. It would allow for cases when a subordinate calls his boss "Sir", but a partner calls him by his name for example. We could weight the personal classifier by how much training examples we had for it.