4 ms·
The problem is making a filter than works for everyone. A 50 something conservative christian woman probably has different ideas about what constitutes as accep
by jiggy2011 13y ago
The problem is making a filter than works for everyone. A 50 something conservative christian woman probably has different ideas about what constitutes as acceptable message to a 20 something liberal woman looking for a more casual fling.
- kaybe 13y agoWell, there's always filter settings.. let everyone rate a few messages into acceptable or not and go from there. (Maybe go from real inbox messages in case one wants to minimize additional exposure.)
- jiggy2011 13y agoYou need a pretty large corpus to train a filter effectively, so probably not practical to train on a per-user basis.
- loup-vaillant 13y agoThe filter doesn't need to be accurate right away to be effective. Also, you could design a training set that gives more information than something made up of random messages. Anyway, if the filter weeds out most spam without getting too much false positives, that's a win. Then, as you mark more messages as "inappropriate", you get less and less such messages.
- kaybe 13y agoAh, that's what I meant.. you don't use those user-classified mails only for training, but you use them to gauge the level they want their filter set to. Then you set the preexisting filter to that, which you have designed using a bigger training set.
- jiggy2011 13y agoAs anybody who's ever tried to train an email spam filter from scratch can tell you , the number of messages you need to classify in order to be useful is huge. With email spam, everyone can more or less agree on what it looks like so you can reuse other people's training data but by deciding on what is a "bad" online dating message is much more complicated.
- loup-vaillant 13y agoOkay.