Windows 2020 Lifestreams and Objectlens's

Meltsner, Kenneth
Mon, 7 Jan 2002 10:18:13 -0500

I looked at Scopeware back when it was Mirrorworlds (~ 3-4 years ago). It had an incredibly cool Javascript/DHMTL interface and the stream notion was quite good -- throwing everything into one big pile and sorting it by the metadata was, after some experimentation, felt to be superior in most cases to file hierarchies.

The argument at the time was that people never organize files consistently into useful hierarchies, and so it was useless to try to maintain them.  I'd argue with "never" -- people frequently organize files into hierarchies that are useful to *them* -- but given the general case that the organizational scheme has to be useful and understandable to *others*, I would still agree with the MW/SW approach of not bothering with directory trees.  Filtering on content or meta-data, plus sorting by time, is a good alternative although not a perfect one.

The big problem that we had was that there was no good way to "paper-clip" together documents that needed to be kept together.  For example, you might want some sort of grouping that associated chapters in a book together.  The query language -- just Verity's full text/field query language -- wasn't rich enough to be able to create streams that captured all of the associations that our users needed.

In addition, it is often useful to be able to find the neighborhood of a document, to look at the other documents that someone felt belonged in the same folder.  The notion of going up in a directory tree to find related categories or more general documents is also useful, and lost in the stream-based approach.  Going from a filter that selects a dozen or so documents to one that selects a larger, more general set is essentially impossible for most users.  Going from a filter for a dozen docs to a filter for a "sibling" set of docs is probably impossible for nearly everyone.  In a directory tree, we can, at least, move upwards to generalize (on one dimension at least), and back down again to another group without much effort.

It's possible that some of this capability could be furnished with the grouping tools in Vivisimo, for example, which can construct ad hoc categories rapidly.  Given a too general stream query, a group system could cluster it in several ways.  Combining that with text query generalization heuristics (substituting thesaurus words, eliminating <AND> terms, etc.), the system could automatically find queries for sibling sets, I suppose.

I'm still of the opinion that you need both arbitrary associations (file folders) where people decide what files belong together and the query-based associations where files are grouped by text or metadata values.  Paul Dourish, who's joined UCI's faculty now, did some good work in this area with the "Placeless Documents" group at Xerox PARC.  The good work, by the way, is partly protected by a group of patents that try to be quite broad in their coverage.

Ken Meltsner