Actually I believe de-normalising the tags table might be a better way forward, depending on scale.
This way, the tags table simply has tagid, itemid, tagname.
You'll get duplicate tagnames, but it makes adding/removing/editing tags for specific items MUCH more simple. You don't have to create a new tag, remove the allocation of the old one and re-allocate a new one, you just edit the tagname.
For displaying a list of tags, you simply use DISTINCT or GROUP BY, and of course you can count how many times a tag is used easily, too.