It all starts with the data…There are 2 approaches to collecting data on one’s interests, influences, and tastes: Implicit & Explicit. Companies such as Hunch and Nara have all attempted to address this by asking their users to share interests explicitly, with the incentive that users will get relevant and accurate choices in return as they use the application.
However, this approach has some inherent flaws. Why would I tell a choice engine that I like Chinese Vegetarian food, only for it to suggest the closest Chinese restaurant soon after? If I mention that I like Asian food in general, would I get suggestions for Korean, which is too exotic for my taste buds? Excerpts from the previous post.
2/ Signals Vs. Noise
“A lie gets halfway around the world before the truth has a chance to get its pants on.” ~ Winston Churchill
Every action that I take is a ‘signal’ of what I want, or what I do, and almost every signal in the digital world has its fair share of noise, and more so on the Internet Highway, though the noise levels and types can vary greatly. For example, my comments “LOL” or “OMG” are extremely noisy, as are my “likes” of random photos and comments.
However, the importance and relevancy of “noise” can be closely correlated to the amount of effort it takes for a consumer to generate that data. The more time and engaged I am in a particular activity, the more accurate the data I generate will probably be.
But the degree of correlation varies depending on the individual person’s behavioural profile. If you take the time to upload a video of yourself sky diving, that’s a strong signal. If you simply “Like” someone’s sky diving photo, it may be that you like the person, or you like sky diving, or you are simply participating in the conversation or seeking attention.
Some sources of data will prove to be more valuable than others, e.g. a simple “Like” on one a product is worth much less than multiple / repeated check-ins at a particular venue. Similarly, “Likes” of brand pages, are probably more reliable interest indicators, and loyalty and transactions / purchases even stronger signals.
Complex algorithms, ontologies, and machine-learning based classifiers are needed to separate noise patterns and spam-like comments, from genuine signals. This is where the complexity arises, and the signal strength calibration algorithm has to then be customized according to each person’s behavioural patterns, lifestyle / context, and transactions.
The beauty is that you don’t have to get it right the first time, since, like any neural network, the Choice Engine will improve & learn with usage over time.
NEXT: Interest Graphs to Choice Engine: 5 critical components to succeed: The Taste Graph [Part 4]