Suggest This, Google « neoSprockets

Suggest This, Google

I knew it.

Google Suggest is quickly becoming my favorite Google tool.

Start typing a string into the Google search box and a cool list of AJAX-powered predicted finished queries are presented to you. “4 9 e” – ah, here it is “49ers schedule”. Very helpful. It seems simple and innocent enough.

But try typing “Christianity is ” or “Judaism is ” and you’ll see a potential problem. Offended much?

So why would Google suggest that “Christianity is bullshit” or that “Chinese people eat babies”? Well, technically they aren’t. YOU are.

Google keeps its Suggest algorithm under strict lock and key, however hints from Google insiders reveal that the suggestions are naturally what you’d expect – they are largely dependent on a search string’s popularity. Those offensive and juvenile suggestions exist because normal people like you and me have actually typed those exact words into Google. And it’s not just one person – tons of people are using that search phrase!

It’s hard to believe though, isn’t it? Are there really that many people searching for “jon stewart is a douchebag”?

Sprocket Ad

How to be Cool

The issue here is the concept of “trending”, where something’s popularity is ironically a function of its own popularity.

You must have experienced this countless times in your own life: Paris Hilton, The Jonas Brothers, Twilight. An item gets exposure by being popular, and more exposure leads to more popularity. It’s a frustrating cycle, especially because the real “value” of the item is typically lost in the mix. Does Paris Hilton have “value”? Are the Jonas Brothers more talented than any other garage band on your block? Is the Twilight story anything new? No! But they’re popular! Why? Because they’re popular! (See my TrendSim post for a cool simulation of this effect).

So how does this relate to Google Suggest? Follow me:

By including a certain search string in the suggestions, Google is increasing that particular string’s exposure. People are now more likely to click the suggestion, even if it’s not what they were originally looking for. This increases the string’s popularity, and cements its justifiable exposure. A search suggestion stays popular BECAUSE it is popular. The problem with this system is that queries of little “value” get sucked into this engine. This is why we see useless suggestions like “neil armstrong is my uncle” and “jackie chan is dead”.

Updating the Google Suggest Algorithm

It’s a tricky problem.

If they aren’t already, Google should strive to improve the Suggest algorithm to combat this “trending” effect. Here’s one way:

Each new search string receives “popularity points” each time that they are a fully 100% text entered search. “Popularity points” decay over time, so to maintain growth, a string must be frequently searched for. If a defined point threshold is reached, a string may start to appear as a suggestion to all users. Once popular, the search string subtracts “popularity points” each time that it is successfully clicked via suggestion. Points can still be added however for natural searches – those where a suggestion is NOT clicked.

This algorithm becomes a self-regulating control system. Popular queries naturally converge to the threshold, able to fall of face of the Internet if they are not readily maintained. It’s not perfect – people might see a popular suggestion and still type it without clicking the suggestion. But it should at least help.

Meanwhile, relax and enjoy the mayhem that Google Suggest brings. Did you know “Google is Skynet” and “Obama is literally Hitler”?