[Standards-JIG] more SPIM JEPs

Tijl Houtbeckers thoutbeckers at splendo.com
Thu Sep 29 18:20:19 UTC 2005

On Thu, 29 Sep 2005 14:10:28 +0200, Hal Rottenberg <halr9000 at gmail.com>  

> On 9/29/05, Tijl Houtbeckers <thoutbeckers at splendo.com> wrote:
>> On Thu, 29 Sep 2005 11:40:38 +0200, Ian Paterson
>> <ian.paterson at clientside.co.uk> wrote:
>> > IMHO the SPIM issue is fundamental to the long-term success of XMPP  
>> (as
>> > the miserable death of email is currently proving).
> Not that I disagree that email is a cesspool of spam, but the burden
> is reduced to neglible levels with a properly administered spam
> deterrent on the server.

In practise, false positives are a big problem. Many companies are even  
moving as much away from email as possible for customer relations because  
their mails get filtered out either at the server or the client. Though in  
email, server can mark messages as SPAM before sending it to the client  
(using several different methods) the client has no way of giving feedback  
wether that message was, or was not SPAM. For example very conistently all  
short messages I send to an @yahoo.com address end up in the junk folder  

In my opinion, this has made email, which used to be quite reliable (more  
so than XMPP) unreliable for sending messages to new contacts. Even  
existing contacts that have serverside filtering are unreliable (the  
server does not know your "whitelist" and can't trust the sender is the  
actual sender). That's somewhat better in XMPP, cause we have a "sort of"  
addressbook, but as pointed out not all your regular contacts need to be  
in your roster per se.

>> As I mentioned way in the beginning of this discussion, I think there
>> should be a way for servers to make an assesment of stanzas, as to the
>> propabilities wether they are spim or not. I think the key to effective,
> And so, analogous to email, whether or not we have quality deterrents
> built into the protocol, the server can always intercept every stanza
> and analyze it using a modification of tools that exist today, such as
> Spam Assassin.

Well, I haven't done any scientific study on this, but with most client  
side SPAM filters I've used (I'd say about 80% of my non-mailinglist email  
is SPAM) the ONLY false positives they make after a while are on short  
small messages. Considering how baysian spam filtering works (though most  
filters use other criteria too) that's not surprising.

More information about the Standards mailing list