Mining the AOL Search Data

Mining The AOL Search Data

By: Mr. Frog
2006-08-29

AOL's release of a treasure trove of search query data gathered from 650,000 of its users has generated a lot of hand-wringing and hair-pulling.

Privacy advocates are angry, lawyers are filing suits, and three AOL staffers have already been fired (including AOLs CTO) over what AOLs senior management has admitted was a screw-up.

It's impossible to applaud the deliberate release of such data, but as a search marketer, it's also impossible to ignore it, because within the 20,000 million queries made by 650,000 searchers there are many fascinating insights into search behavior that will be of great interest to search marketers. Several important findings have already emerged from the study which accompanied the release of this data, titled A Picture of Search, authored by AOL researchers Greg Pass and Abdur Chowdhury, and consultant Cayley Torgeson. They include the following:

1. Click-Through Behavior and SERP Position. The #1 position on AOLs SERPs accounted for 45 percent of total click-throughs. The #2 spot received just 13 percent of clicks. Why such a disparity? The study posits that user habits, branded experiences, page layouts, surrogate quality all combine to create a discontinuity between users perceptions of utility and traditional measures of relevance. This finding closely correlates with eye-tracking studies performed by Did-it and Enquiro (http://www.prweb.com/releases/2005/3/prweb213516.htm), so it may not exactly be surprising, but it reinforces the importance of achieving top rankings for all significant keywords driving your business. Like it or not, theres no substitute for being #1.

2. Domain Name Searches. A high percentage of searches aren't for topics, but for domain names. (Rand Fishkin, of SEOMoz.org, estimates this number at 21 percent). In many instances, users seem to regard the search query box as a substitute for the browser address field. Naturally, if you're doing keyword discovery, you don't want to overlook domain-oriented keywords.

3. Misspellings and Mashed Queries. The number of misspelled, incomplete, or syntactically incorrect queries is very high, although no study has quantified it yet. Make sure you've done your due diligence in terms of discovering all the possible ways that such queries can be mashed (there are obviously a lot of incorrect permutations!).

4. Query Reformulations. 28 percent of queries were reformulations of prior queries. Among these, queries were reformulated an average of 2.6 times. Reformulations can occur when a user realizes that he/she has mistyped an initial query and receives irrelevant responses on a SERP, or when he/she adds terms to narrow down such queries. By studying the reformulation process, you can gain insights into how queries are refined as the user moves through the decision funnel.

Several sites have already popped up offering different ways of viewing the AOL data. Among the most interesting are:

1. AOLPsycho (http://www.aolpsycho.com)
This site leverages the power of social tagging in an attempt to create descriptive profiles of each of the 650,000 AOL users included in the database. Many of the profiles are highly cursory, but for users with very long query strings, its possible to draw valid conclusions about the psychology of AOL users. Warning: some of the search query data reveals troubled individuals with exotic tastes which may be offensive.

2. SEOSleuth (http://www.seosleuth.com/site/)
This interesting site lets you directly compare SERP positional results with click-through rates on a keyword basis. It also lets you search by domain name, inspect the amount of traffic each site got, and which keywords were used to access the domain. If youre doing keyword discovery, this site can be invaluable and may reveal keyword permutations (including misspellings) that you might otherwise have a hard time thinking up on your own. SEOSleuth also provides limited dayparting information for each query (time of day but not day of week).

Other sites which let you browse the AOL search data include:

http://www.aolsearchdatabase.com
http://www.datablunder.com
http://www.frogspy.com

Some researchers have stated that they consider the AOL data tainted and refuse to perform any analyses on it. And its certainly true that AOL should have done a much better job scrubbing any personally identifiable data from the query database (such as vanity searches for ones own name or queries which included a credit card or Social Security Number). Still, there are so many useful insights in this data that search marketers will find it almost irresistible to use this data to inform the way they conduct search campaigns. Now that this data storehouse is in the public domain, Id expect a lot of additional compelling insights to emerge in the near future.

Actually, given the notoriety of this data perhaps the level of study on this data will exceed what might have been done had the data leak not been controversial. It may actually be in the best interest of all the search engines and the industry for scrubbed data like this to be released to academia and selected research organizations.

Tag: AOL

Add to

Del.icio.us |

Digg |

Yahoo! My Web |

Furl

Have a bookmark! -

View All Articles by Mr. Frog

About the Author:
Mr. Frog is a leading Search industry visionary. Mr. Frog is a member of the Did-it Search Marketing team which accompanies him to most major marketing conferences.