RecSys Strands Startup Prize October 24, 2008Posted by Andre Vellino in Collaborative filtering, Recommender service.
I’m quite enjoying myself at Recommender Systems 2008 in Lausanne. It’s grown by 50% from last year and the papers are excellent. Continuing from last year’s tradition, there’s also a significant “commercial” strand at the conference and this afternoon has been devoted to 15 minute “elevator pitches” from 4 recommender startup companies who are competing for the Strands $100,000 startup prize.
7 years after the imGenie startup that I was involved with for a while, Recoon (web site not up yet) has come up with exactly the same idea for location-based mobile recommendation of restaurants and such like. This time — now that we have the iPhone and Google Phone — I predict this will work!
If I were to offer my submission for a (now topical) application for recommender systems it would be a stock recommender that uses Mutual Fund holdings as stock ratings. Here’s how it works (if you implement it and make a buck, let me know).
Harvest the stock-holdings of mutual fund companies in different categories (“growth” / “dividend” / “energy” / “technology” / “financials”, etc.) and order these holdings by percentage held by the mutual fund. Calculate the performance rating for the mutual fund algorithmically and use both these sets of data points to populate a ratings matrix for funds-stocks (analogous to user-items).
Now, for a given user that has a portfolio of stocks (or mutual funds, for that matter), find the neighbourhood of mutual-fund profiles that are most similar and use those with a Collaborative Filtering recommender to offer the user stock recommendations.