Seth wrote:
Jose-Marcio Martins da Cruz
<Jose-Marcio(_dot_)Martins(_at_)mines-paristech(_dot_)fr>
wrote:
Quoting a research paper from Gordon Cormack [1] :
"When explicitly asked to classify messages, human subjects have
been reported to exhibit error rates of 3%-7% [30, 16]. Tacitly
derived labels, such as those obtained from a “report spam”
button, where it is assumed that unreported messages are ham, may
have even higher rates."
The error rates differ when the user selects a range of messages for
the button and when the user selects single messages; the former
generates much higher error rates.
Probably. Do you have references about that ? Comparative error estimates ?
--
_______________________________________________
Asrg mailing list
Asrg(_at_)irtf(_dot_)org
http://www.irtf.org/mailman/listinfo/asrg