Right now, there’s limited information on reputable news sites regarding the earthquake/tsunami that occurred south of Samoa yesterday. To find more, I turned to Twitter (and its open source equivalent identi.ca). Disseminating live news from witnesses, moments after a major event, is the most compelling feature of such services, but both failed me. Not in the fail whale sense. Both sites were technically operating at 100% normal status. It was the content.

For example: using search.twitter.com I get the folowing results for samoa

1. Hypeflash: Alao Siva-American samoa flag day 2008 http://tinyurl.com/[REMOVED]
2. Hypeflash: Bizarre Foods SAMOA 4 – http://tinyurl.com/[REMOVED]
3. lisa1248: Get free medicines and sample for free http://bit.ly/[REMOVED] #everlastingsong Goodnight Google Wave #HealPhilippines Kraft Life Motto Samoa
4. dixiefs51: after Manila, Typhoon Ketsana ( ondoy) has killed people in Vietnam then a tsunami hit Samoa.
5. jessecardol3e: American Samoa | got a cam? wanna go one on one? http://wowurl.com/[REMOVED]

Only one of the most recent five tweets is of any relevance, and even that is just passing on the story, not adding any detail. The rest all link off to sites selling medicines or other vices. I’ve got to look harder to find more information, but wading through an 80:20 spam ratio is not convenient. When looking for timely information, this kind of noise can only lead to the source being dropped from the search.

Live news from witnesses, moments after a major event, is Twitter’s most compelling virtue. The problem is, ne’er-do-wells looking to make a quick buck are now all over it like a bad rash.

Now, it might be possible to utilise an n-degrees of separation model for searches. I might do a search for samoa, and restrict results to those written by people withing 4 degrees of separation from me: however, such a system is almost certain to be useless, because there’s no guaranteeing that my expended network is affected by and recording the event.

A more fine grained model of trust might be a solution here. I don’t necessarily know everyone I follow on twitter, but I do know who my friends are and who I trust. If there was a way to put a figure on that trust then rather than the binary degrees-of-separation model, a probabilistic model could be used where I ask for search results within a trust threshold.