Search tagging

Last year at work we were kicking around a product idea (codename: Shreveport) that would log users' outbound search terms as well as the results they select. And through some algorithmic mojo these things would create ad hoc communities of interest, where you could find people with similar interests based on what they searched for. It would also enhance recovery, findability and discoverability of information by "attaching" search terms to URLs (tagging, in essence, though I didn't think of it that way at the time).

This was in the pre-folksonomy days, so we were thinking of Shreveport as a niche knowledge management product. In the meantime, del.icio.us and Flickr did similar things using tags. And Eurekster had some of the social software functionality we were imagining.

We didn't have capacity to do product development at the time, so the idea was shelved. But I've been revisiting Shreveport lately and it remains interesting.

Shreveport is basically a search engine with some social software concepts layered on top. Here are some of its core concepts:

• Search terms are tags on an URL. Shreveport associates tags with URLs based on clickthroughs.
• Search history is shared. Search terms and selected results are shared in the same way del.icio.us shares tags and URLs. (Obviously, we'd thought about an opt-out feature for some searches.... you know, "athletes foot remedies" and the like.)
• Search terms and results selection help improve search results. Part of our largely hypothetical algorithmic mojo engine was a way to use improve results by tracking which links were selected for each query. (I wonder if anyone's doing this now?)
• Exploration and recommendations. Users can explore tags, URLs, users and their visited results. For each search they see weighted recommendations ("People who searched for 'celiac disease' also searched for...") and recommended links based on others' searches.
• Ad hoc social networks. The community aspects of Shreveport were completely ad hoc, based only on search terms. No adding people as contacts or joining networks. Clearly this feature works better for populations with a strong shared vocabulary. (This is similar to what del.icio.us does with tags, but at the time it seemed much more radical.)
• Presence. The original Shreveport concept incorporated presence to encourage direct interaction between users.

One of the things I like about Shreveport is how it doesn't require any new information. Not even tags. It leverages data that users already reveal--search terms and results selection.

It also ties folksonomies seamlessly into the search process. Instead of picking up where search leaves off, they are one and the same. (There is a certain presumption of findability involved in del.icio.us or furl.) James Melzer mentions a similar idea in his post on implicit folksonomy building. Shreveport is both implicit and effortless.

I know I'll never build Shreveport (indeed, I tend to underestimate the effort involved in the algorithmic mojo part). But looking around at the major search offerings, I think A9 is best positioned to make search-folksonomy play like this. And Amazon already knows the benefits of social findability through its Listmania and collaborative filtering features.