I've been working on a really neat problem this week for KType - how to compare phrases and find the best matching ones. If a user types "wnv u wnt", I want to autosuggest "whenever you want." Every computer and phone has spell-check to fix spellings but there is no established algorithm to convert abbreviated speech into regular English, especially for multi-word phrases. Certainly some words are easy to guess like wnv = whenever but what should "smtg" be? Something? Smoothing? Smothering? Smelting? Well that depends on a lot of factors like the popularity of the word, the subject matter, the words that come before, and most importantly the actual context. "I wnt to tl u smtg" most probably means "I want to tell you something."

I've been reading up on algorithms like Bitap, Levenshtein, Bloom filter all day and haven't come across any algorithm that I can directly use. Looks like I'll have to figure this problem out on my own. Fun times!

Add a Comment