Search Journal

When I first decided to build this search index I figured my days of traditional bookmarking were well and truly over. I was wrong. Despite the advantages of searchmarking over bookmarking, there are definitely times where a bookmark is entirely appropriate and does a bang up job.

Two months after deleting all bookmarks from my machine at work, I now have a total of 12 bookmarks there. Without exception they are links to corporate resources I would not have access to without my job: the corporate Internet, web mail client, internal sites, and private research & search resources. I have no need to remotely search these resources as they all provide their own search and/or browse tools, most of which are adequate for the things I need to do. They are links to resources I use frequently enough to want convenient access to them, but infrequently enough for me to commit the URLs to memory.

"... allow us to reconnect to the original bookmark metaphor: a bookmark of the noncyber variety, if we recall, is used to mark a place within a book so that we can continue on from that point at some later time ... But never did traditional media coax us to purchase a product to manage our bookmarks - with the possible exception of the wastebasket."

-- from The Bookmark Less Traveled (Wired, April 1996) by The Sucksters

The bookmark metaphor has always struck me as odd. It seems to suggest an interruption to something linear, like having to halt partway through a good novel just to answer a telemarketers phone call. It also suggests a certain degree of negligence. If I had 300 novels kicking around my apartment, each only partially finished and littered with little scraps of paper, I think I'd find that a bit disconcerting.

Calling them "favourites" is a bit better, but not by much. It seems to suggest a place you'd like to return to, a place that holds some special and personal meaning. Try as I might, I just can't think of a inter-office phone list as a place I'd like to return to, or a web application for filing time sheets as having special and personal meaning. Truth be told, these are places I need to return to, and that have personal meaning to me only to that extent.

The best metaphor I've come up with so far occurred to me while camping in Kluane National Park a few years ago. It's like marking waypoints on a map, or GPS device. You are marking places of distinction, convenience, and safety scattered across a geography of which most of what you know is based on where you've been, and what was there.

About a month ago I was trying to find the URL for an article I first read back in mid-2000 and was about to recommend to a friend: Style-free XSLT Style Sheets by Eric van der Vlist. A quick search for Eric van der Vlist dug up the article in question, but also unearthed another article I had indexed: Building a Semantic Web Site. I hadn't realized these two articles were by the same author until right that second.

I decided it would be interesting to see the results for other authors, and tried some random searches: Peter Van Dijck, D. Keith Robinson, Jef Raskin, Niall Guerin, Clay Shirky, and Tim Berners-Lee to name but a few. It was interesting how often I have bookmarked multiple articles by the same author without noticing the connection.

Of course, I'm not really performing a search for documents by "Tim Berners-Lee" (for example), I'm actually searching for documents containing the phrase "Tim Berners-Lee". Because of this the search results will contain documents that cite Tim Berners-Lee as well as those by him. In fact, with an index of 2006 resources, a query for "Tim Berners-Lee" produced 34 results, as did a query for "by Tim Berners-Lee". No help there. Preferable to manipulating the query would be the ability to establish context, say in one of the following ways:

• Search for "Tim Berners-Lee" in the content of an HTML or XHTML <address> element.
• Search for "Tim Berners-Lee" in the content of an HTML or XHTML author and/or copyright <meta> element.
• Search for "Tim Berners-Lee" in the content of a creator, contributor, and/or rights element from the Dublin Core Element Set.

This wouldn't be such a difficult thing to arrange, but it does open an a wide world of other options. Why just authordiving? Why not be able to refine searches to various elements that define titles, publication dates, block quotations, citations, or term definitions? Version 2.0 perhaps.

Aside: In writing all of the the above I was once again prompted to search for Eric van der Vlist. While the results did include references to Style-free XSLT Style Sheets and Building a Semantic Web Site, there was also a new result: the XMLhack.com home page. It turns out that xmlhack.com was promoting the publication of A RDDL repository of core datatypes by Eric van der Vlist, published March 26th, 2004. I figured it couldn't hurt to index it.

The 2,000th resource added to the search index was: 3.4 The Connection Between Eye-Gaze Pattern and Interest from Eye Controlled Media: Present and Future State.

In celebration I rebuilt the entire search index. The reindexing process took 126 minutes and consumed about 4 MB of bandwidth according to my hosting provider. The resulting index is an impressive 12,600 KB in size. Quite a bit larger than your standard bookmark file to be sure, but still a small price to pay so far for the advantages that searchmarking has over traditional bookmarking.

Reindexing shaved 193 KB from the size of the index. 13 resources failed to be updated and 30 resources were removed altogether.

In hindsite, what I was looking for were the following two resources: www.pixy.cz/apps/barvy/index-en.html and more.btexact.com/people/rigdence/colours/. The process of finding them in a search index of 1,867 resouces was not as easy as I would have hoped.

Trying to cultivate the best set of results, I searched with the following sequence of query patterns:

• colour and blindness producing 3 results but no link to either pixy.cz or more.btexact.com
• colour or blindness producing 42 results but no link to either pixy.cz or more.btexact.com
• color, colour, or blindness producing 293 results with a link to more.btexact.com as result #21 and a link to pixy.cz as result #23.
• "colour blindness" or "color blindness" producing 6 results but no link to either pixy.cz or more.btexact.com
• "colour blindness" or "color blindness" or "color scheme" or "colour scheme" producing 9 results with a link to pixy.cz as result #3.
• safe or color or colour or blindness producing 370 results with a link to more.btexact.com as result #9 and a link to pixy.cz as result #25.

There are several concurrent problems of consequence at work here:

• The relevance or quality of online tools are rarely indicated by the number of times specific keyword patterns appear on the page.
• Abstracts and summaries of resources cannot represent the relevance or quality of the resource they reference.
• Differences in the spelling of words between countries and dialects.
• The search engine's interpretation of each query probably differs from my own. Jack Muramatsu and Wanda Pratt wrote some insightful words on this in "Transparent Queries: Investigating Users' Mental Models of Search Engines"

Of course, it's exactly these kind of shenanigans that prompted google to develop their PageRank Technology (and propelled it to the most popular search engine in the world). Alas, however, this idea holds little hope for this search engine as the act of evaluating link relationships en masse would take exponentially more server resources than the act of indexing the resources themselves. Furthermore, the very public nature of these comparisons might serve to dillute the personal nature of this index.