Talk:Bayesian message classification: Difference between revisions
imported>Catherine Woodgold (It's more complicated than that.) |
imported>Subpagination Bot m (Add {{subpages}} and remove checklist (details)) |
||
| (One intermediate revision by one other user not shown) | |||
| Line 1: | Line 1: | ||
{{ | {{subpages}} | ||
}} | |||
== It's more complicated than that. == | == It's more complicated than that. == | ||
This article gives the impression that Bayesian spam filtering is done in a particular way, i.e. by treating probabilities of each word independently. That is not the only possible way to do Bayesian spam filtering, and I don't think it's the way it's usually (or always) done. Another way is to look at probabilities of phrases. Yet another way is to look at probabilities of certain combinations of words (regardless of where in the article the word appears). For example, the word "interest" might not by itself increase the spam score (or not much), but if it appears in the same message as "mortgage" and "house" it might add significantly to the probability of the message being classified as being about mortgages, and then get a Bayesian spam score based on the user's previous reactions to other messages about mortgages. In other words, it can be done in two steps, using Bayes' theorem at each step. --[[User:Catherine Woodgold|Catherine Woodgold]] 21:22, 2 May 2007 (CDT) | This article gives the impression that Bayesian spam filtering is done in a particular way, i.e. by treating probabilities of each word independently. That is not the only possible way to do Bayesian spam filtering, and I don't think it's the way it's usually (or always) done. Another way is to look at probabilities of phrases. Yet another way is to look at probabilities of certain combinations of words (regardless of where in the article the word appears). For example, the word "interest" might not by itself increase the spam score (or not much), but if it appears in the same message as "mortgage" and "house" it might add significantly to the probability of the message being classified as being about mortgages, and then get a Bayesian spam score based on the user's previous reactions to other messages about mortgages. In other words, it can be done in two steps, using Bayes' theorem at each step. --[[User:Catherine Woodgold|Catherine Woodgold]] 21:22, 2 May 2007 (CDT) | ||
== Why is this article marked external? == | |||
If it's from another source, then what is it? [[User:Greg Woodhouse|Greg Woodhouse]] 12:10, 30 June 2007 (CDT) | |||
Latest revision as of 03:37, 25 September 2007
It's more complicated than that.
This article gives the impression that Bayesian spam filtering is done in a particular way, i.e. by treating probabilities of each word independently. That is not the only possible way to do Bayesian spam filtering, and I don't think it's the way it's usually (or always) done. Another way is to look at probabilities of phrases. Yet another way is to look at probabilities of certain combinations of words (regardless of where in the article the word appears). For example, the word "interest" might not by itself increase the spam score (or not much), but if it appears in the same message as "mortgage" and "house" it might add significantly to the probability of the message being classified as being about mortgages, and then get a Bayesian spam score based on the user's previous reactions to other messages about mortgages. In other words, it can be done in two steps, using Bayes' theorem at each step. --Catherine Woodgold 21:22, 2 May 2007 (CDT)
Why is this article marked external?
If it's from another source, then what is it? Greg Woodhouse 12:10, 30 June 2007 (CDT)