Frrole is an interesting startup that is trying to make sense from Twitter noise by creating city-based social newspapers from Twitter feeds.
That is, the social newspaper for say, Bangalore city will have all the relevant newsy tweets (especially for categories like Entertainments/Sports and Travel), tweets that talk about events and also, tweets those mention deals.
Talking about the algorithm, Frrole founder Gurjit Sidhu shares the following:
Speaking at a high level, the algorithms have three components. Once the “in city” and “about city” data imports from twitter are synced up, the first set of algorithms is built around metadata analysis. Metadata information around language, location, type of tweet etc. is made available by twitter and we use that for analyzing the streams and reducing the data sets. In between metadata algorithms, we do volumetric analysis that lets us do things like counting the no of tweets, retweets for each, factor in our rules for the cities etc.
The final component includes a textual/NLP analysis where we parse the data sets to infer the nature of tweets. Keywords are an easy but low quality way, syntactical construction of sentences is tougher but a higher quality way of inferring that information.That is one reason why we are limited to English language, but that itself presents a pretty big market opportunity. We do not use any 3rd party NLP libraries, our needs are very specific, so we have just built our own processing engine.
In addition we also try to use Bayes networks to determine probabilities of a tweet being relevant/irrelevant and then categorize it. The results of the processing engine is served by a services layer which has caching layer(to handle load .. for the IF case:)). Our plan is to expose these services as well as build android app/iPhone apps using these services .
–
The only and probably, the only question that the team need to answer is whether Frrole is dependable or not. That is, it has to work as-expected (i.e. like a social newspaper) or there is nothing for people to even revisit the site (already, we are going through information overload when it comes to social media). In its current form, the product is doing a great job of filtering categories, but I won’t say its 100% accurate (shows random/Delhi events for Bangalore city). Least the team can do is to add a ’Flag this’ feature which will power users to flag a tweet as inappropriate.
Do give Frrole a spin and share your comment. Does it has the stickiness needed to become a destination site? Do share your feedback with the startup.
Aside, is it just me or do you think the name is too tongue-twisting (too difficult to type?).
– Follow @pluggdin on Twitter for some ‘startup noise’.
