PredictionPlatform.com

I often find myself wondering about what factors major social networking sites use to recommend users to interact with (friend, follow, connect, etc.).

Facebook appears to use the number of friends in common as the major factor. Personally, I find these recommendations are qualitatively ok, as Facebook friends are usually friends of some sort, the data has decent quality.

Twitter uses a similar strategy, but I believe they may also include general popularity of a user when making a recommendation. Twitter ranks the worst in recommendation quality in my books. It appears negative feedback in the form of removing recommendations is ignored and I am still prompted to follow celebrities more often than not.

LinkedIn's "People You May Know" scares me with its high quality of recommendations. Industry, interests, and geography seem to be in play besides the number of common connections. Strangely, names of people I know from the past in school are listed... though they aren't actually the same person that I know. I wouldn't be surprised if LinkedIn uses outside data sources as food for their algorithms.

I'll be on the look out for more empirical examinations of these algorithms and continue to wonder if my interest in them actually skews my evaluation of their quality.

I think predictionplatform.com is a decent domain name to front an idea I have been dabbling with for a few years. A general purpose prediction web service can power tools for business, government, and individuals; new web 3.0 (whatever you want to call it) data-driven tools need to harness the potential of artificial intelligence (machine learning) to discover patterns and connections in the mountains of data that we are producing. The skills and expertise needed to create the software to expose these valuable data insights aren't easy to find though, so a generalized prediction service or platform may be the answer for next generation applications that need to dig deeper into social graphs, transactions, and habits.

So I bought this domain in 2009 and have been researching and experimenting since. A prediction platform is a huge undertaking. So to take some baby steps toward something concrete, I've turned the domain "on" and will start indexing relevant technology and developments, ideas, and applications. Hopefully the practice of getting my ideas and thoughts out of my head and onto the web will help nurture them.

As a hobby, I expect it will take a few more years before I have something worth calling a product. Until then, I hope to share my progress and research and connect with anyone else with an interest in the area.