Nemáte pravdu, děláte nesmysly na Blog.cz

Principle of tagging which I am sorting to "first steps to semantic web" is simple. We can just use actual model of grab-it services as our base. Let's say anyone can tag every page he find. We get 1xN database of pages and tags. But what we do now is something possibly new: if tag set is searched, we select pages where is any of tags searched, we order all pages by tag intersection to query primarily, and then we sort it secondary by count of taggers signing one document more similar way.

It is also easy to implement neural nets and some semantic relation-dictionaries, cathegory dictionaries also, which may extend this low level "statistical computing on tag data" to some AI level, for which is no problem to find "Latest HIV vaccination was not unsuccessful at clinic tests" news when you only search "Medicine news" and noone tagged this article with any of your query tags. But basically making summations from many tags to single page in combination with tagging groups should be bullet-proof relatively-semantic improvement to web searching.

The good thing in having tags is its "word relation" and "notion cathegory" implicit content. For cathegory is needed only option to write tags in "A/B C D" form.

Let's resume two actual types of tagging: rdf and grab-it web services. Rdf is absolute way of tagging, what means that every document has one meaning, and this meaning is hoped to be loosles requackquacked in rdf "header" of file. Also rdf tagger is hoped to be the best "understander" and also best "requackquacker". On the other side grab-it services are working with relativistic view of document, but are not thought to be semantic, what leads to quick and improper or very simplifiing tagging. On the other side these services are working with local-relativistic tag value. It means that there is no synthetic tag value for every document, but only many tag evidences for any.