Measuring "Irish-ness" with the Murphy Index :shamrock:

:bulb: The Idea:

Spam detection can be tricky business! Spammy emails or text messages generally have plenty of words to work with, but what if we only have a couple of words to work with? What if we want to detect whether an account is spammy just by its name?

:gear: The Method:

Turns out, this is just the sort of thing the Naive Bayes algorithm is good at. We can divide the name(s) into ngram tokens and use those tokens to calculate a probability that a name is spam. For instance, if my name is Eric and I am using 3-grams, then the document would become Eri ric. Ngram length is one of the hyperparameters I tuned during model development, which you can review in more detail in this Jupyter notebook.

Based on the excellent work of David Mandell Freeman shared in this paper.

By comparing average log predicted probabilities of name ngrams being Irish to the "benchmark" Irish surname "Murphy", we can create a tongue-in-cheek "Irish-ness" score just in time for St. Patrick's Day!

For example, if the average log predicted probability of the name "Murphy" is -6.58, and the average log predicted probability of "Bailey" is -7.762, then the Murphy Index would be (-6.58/-7.762)*-100 = 84.77. A bonnie lovely Murphy score, I'd say! :)

:iphone: The App:

The app is easy to use: just drop your surname (or your friend's surname!) in the text box, press Enter, and get your Murphy score.

As a bonus, you can see your score stacks up against others who have used the app before you. I thought users would enjoy seeing how their Murphy indices compare to others, so I connected to the Google Sheets API and store each score produced by the model and present the results below the score results. I've gotten a lot of great feedback on it!

Try it out yourself here!

I'd love to hear what you think! Feel free to message me on LinkedIn or submit an issue or pull request here.