I enjoyed this blog post: On Serendipity. Ironically, it was recommended to me, and I am now recommending it!
Serendipity is rarely of use to the asset manager, who wants to find exactly what they expect to find, but is a delight for the consumer or leisure searcher. People sometimes cite serendipity as a being a reason to abandon classification, but in my experience classification often enhances serendipity and can be lost in simple online search systems.
For example, when browsing an alphabetically ordered collection in print, such as an encyclopedia or dictionary, you just can’t help noticing the entries that sit next to the one you were looking for. This can lead you to all sorts of interesting connections – for example, looking up crescendo, I couldn’t help noticing that crepuscular means relating to twilight, and that there is a connection between crepe paper and the crepes you can eat (from the French for “wrinkly”), but crepinette has a different derivation (from the French for “caul”). What was really interesting was the fact that there was no connection, other than an accident of alphabetical order. I wasn’t interested in things crepuscular, or crepes and crepinettes, and I can’t imagine anyone deliberately modelling connections between all these things as “related concepts”.
Wikipedia’s “random article” function is an attempt to generate serendipity alogrithmically. On other sites the “what people are reading/borrowing/watching now” functions use chronological order to throw out unsought items from a collection in the hope that they will be interesting. Twitter’s “trending topics” use a combination of chronological order and statistics on the assumption that what is popular just now is intrinsically interesting. These techniques look for “interestingness” out of what can be calculated and it is easy to see how they work, but the semantic web enthusiasts aim to open up to automated processing the kind of free associative links that human brains are so good at generating.