Most publications on researchr don’t have tags associated to them. This makes tags not really useful for (1) navigating within collections (search results, publication lists like bibliographies, publications from an author or conferences etc).

I think it will only be useful for this job ¹ if all publications are tagged, appropriately. Tagging all publications must not be limited to extraction of terms from titles. Abstracts are more detailed and include most terms that relate to tags. So in order to have useful tagging I think that we first need to have a auto-fetcher for abstracts. Then we must investigate how to discover tags from texts automatically, and how to auto-tag publications based on title and abstract.

I think tags on researchr are currently only useful to filter a collection that is known to be tagged correctly for some essence, probably like the webdsl tag.

Otherwise, filtering on a tag will only give you a (probably) very small subset of the complete set of publications that should have been tagged with the one that is filtered. The user will probably experience this as system returning incomplete results, or the user will even be unaware of the fact that only a subset of tagged publications are presented. Users may loose confidence in the browse functionality of the web site because of this. Also, ordinary users don’t need to bother if “browse results” are complete or not, and why, unless stated explicitly.

I would personally change current tag clouds in one of the following ways:

• remove the tag clouds completely
• keep tag clouds, but make users aware of their limited usefulness (next question: how?)
• (prefer:) try to add abstracts automatically for all publications and auto-tag all publications using texts from title and abstract. This will improve user experience a lot.

Submitted by Elmer van Chastelet on 1 October 2012 at 12:14

I think experienced web users understand that tags are user-generated content and therefore not complete.

I agree that we should try to get abstracts for publications. Since we have DOIs that should not be very hard.