We talked in class about Google Trends, and touched on the idea that Google was increasingly using information collected from users as they searched to improve its engine. It appears that these two things actually go hand in hand, and that Google trends itself is useful user data, although not in the conventional sense.

The way user data is typically used by a search engine involves watching what links people click when they search for a certain query. As a simple example, if everyone who googles ‘stewart’ clicks the fifth link, that’s an indication that the fifth link should be moved up. Conversely, if people googling ‘OED’ don’t click any links, but then immediately search for ‘Oxford English Dictionary,’ that tells us something about what they were looking for.

Google Trends is user data too, in a different sense. Just seeing the popularity of various search terms isn’t terribly helpful to a search engine, unless it can somehow predict when those terms become popular, as in the case of annual holidays and so forth. But the ability to compare, and correlate, different search terms is tremendously powerful.

As an example, look at this. We see a few interesting features of this graph. First, whenever there is a spike in searches for “Barack Obama” there is a corresponding spike in searches for “Obama.” The graphs are incredibly well correlated. Furthermore, as time goes on, searches for just the last name “Obama” become more prevalent than searches for his full name. Note that the graphs still follow each other quite nicely. Finally, almost nobody searches for just the word “Barack.”

There are several conclusions we can draw from this. First, when people search for “Obama” they almost certainly mean the Democratic Senator from Illinois. Whenever there is a jump in searches for one, due probably to a news story or other real world event, the other follows suit. Additionally, Barack Obama has become increasingly associated with just his last name: whether as a result of increased name recognition, or just the perception among users of increased name recognition, more people look for him using just his last name. As a corollary, almost no one identifies him by just his first name, either because they know the query won’t be understood, or because it’s simply more conventional to identify politicians by their last name.

In effect, we can use Google Trends to connect ‘nickname’ searches with a more specific proper name. This pattern is borne out by many other queries, although it’s perhaps easiest to observe in famous people. There are counter examples: Martha Stewart is not strongly correlated with either her first or last name, because they are not ‘nicknames’ for her. On the other end of the spectrum, Cornell is so strongly identified by its ‘nickname’ that there is basically no connection between the search trends for “Cornell” and “Cornell University.”

Given the vast amount of data and processing power at Google’s disposal, it’s possible that these trends could be added to their new feedback powered engine in the future.