This is an old book (really old) published in 2008. I didn't manage to get hold of new data science-related books as they are mostly on loan. However, it's interesting to note that using data to understand people's behavior and intentions had started way long before. It's a good book for those who're starting out; it provides several use cases of internet search data.
I've picked out some quotes that I liked/ could relate to:
- Mining through observed behavior, at times, is like a treasure hunt. It requires both the right-brain skill of seeing patterns in the data and the left-brain ability to formulate a hypothesis and test it with the data at hand. (pg 48)
- In the case of our New Year's resolutions, as in other areas where we seek to better ourselves, the quick, get-it-now conditioning that Internet connectedness has provided us with has left us with a very limited attention span. (...) While our attention span has shortened when it comes to self-improvement, our fascination with the lives of celebrities has grown significantly within the last several years. (pg 84-85)
- (...) we weren't exactly predicting anything with our data; rather, we were each just very fortunate to have a data set that allowed us to see what was already happening in a market- be it housing, employment, consumer spending- before anyone else had the opportunity to see what was happening. It seems like a simple distinction, but there's quite a difference between the art, often perceived as the voodoo, of predictions versus straightforward arbitrage. Predictions involve assumptions, calculations, and, above all, substantial room for error, depending on the precision of your underlying assumptions and calculations. Data arbitrage, on the other hand, is simply taking advantage of the time differential between when people do something on the Internet (such as search for a home for sale, or place their residence up for sale) and when a financial indicator such as existing home sales would show that activity. The gap between the two can be a matter of days or weeks or even months. The greater the better. (pg 156-157)
- (...) data is always right but pitfalls exist in how we interpret it, from gleaning insight from search terms to knowing the difference between search intent and actual behavior. (pg 169)
- While data itself can be the most effective predictor, filtering all that we know about a large sample of Internet users can produce broad results. (...) In some trend predictions, it's best to start with a well-connected, tech-saavy individual, someone who would be voted most likely person on the block to adopt the latest technology - the Early Adopter. (pg 170)