Unless you're already a hopeless Twitter addict, dipping into the service is a headspinning experience. You're supposed to comb through this huge, rambling list of crap? And that's fun?
The solution, of course, would be a system that accurately categorizes tweets, according to topics and what's hot. But so far, nothing has really succeeded in doing so (sure, there's the #hashtag designation that marks topics, but it's still a #%$&@ mess).
Dubbed Eddi, the service has two tiers of grouping. One highlights tweets that you'll probably find interesting, given who's writing them and any overlaps with your own tweeting history. But the second grouping system is really ingenious.
Eddi takes tweets and feeds them into Yahoo!'s search engine; the search engine in turn points Eddi to the topics that are likely contained in the tweet. Eddi then classifies all the tweets it has combed through, and summarizes all the activity in a tag cloud.
The appeal of that approach is that it doesn't rely on semantic, natural language analysis to guess at a tweet's content--that technology would likely get too much wrong to be useful. Rather, it generates tags by tapping the web's collective intelligence, and the inter-relations that humans have already identified--an idea similar in spirit to the pagerank algorithms that have made search engines so indispensable.